On Sunday, the U.S. Department of Commerce took steps to address a potential loophole that has existed for a year, which might have permitted companies to export cutting-edge chips—such as Nvidia's top-tier Rubin and Blackwell processors and AMD's MI350x—to Chinese entities situated outside of China.

This new directive indicates that some of the United States' most advanced AI chips may have been sent to subsidiaries of Chinese AI companies based in locations like Malaysia for nearly a year, countering the U.S. government's broader strategy to restrict the flow of critical semiconductors necessary for developing significant AI technologies in China.

The guidance was made publicly available on the Commerce Department's website over the weekend.

The exact volume of chips exported during this one-year period of regulatory leniency is not clear, but one industry insider with extensive knowledge of supply chains estimates the figures could rise into the hundreds of thousands.

In an unexpected move, the Commerce Department stated it would now enforce licensing requirements for advanced chips traded with entities based in China, regardless of their actual location.

The agency did not provide immediate commentary regarding this new measure, nor did Nvidia or AMD respond to inquiries.

This oversight originated when the Commerce Department announced back in May 2025 that it would not enforce the AI Diffusion rule established just before President Biden's term concluded, which regulated international access to AI-related chips.

Chris McGuire, a technology consultant and former State Department official, commented on social media, deeming the situation to be "a HUGE problem.” He pointed out that this loophole enabled overseas subsidiaries of Chinese firms to procure Nvidia Blackwell chips without the need for a license.

"Chinese companies have been acquiring these chips, very likely in large quantities," McGuire noted.

Interestingly, the updated guidance does not require data centers to cease utilizing the chips or to withdraw services for advanced computing systems like servers.

U.S. moves to stop Nvidia AI chip exports to Chinese companies outside of China.

A dark, moody port warehouse at night: heavy wooden crates stamped with a large Nvidia logo are lined up on a dock, and a silhouetted U.S. customs officer slaps a bold red “STOP” stamp across a shipping label reading “Shanghai Bound.” Behind him on a corrugated-metal wall hangs a single, prominent U.S. flag. In the distance under a lone overhead lamp, a cargo ship loaded with sealed chip containers fades into shadow—dramatic naturalistic lighting casting long, tense silhouettes.

U.S. takes step to halt Nvidia AI chip shipments to Chinese firms outside China - CNBC

- The U.S. Department of Commerce closed a loophole allowing advanced chip exports to China.  
- Advanced AI chips reportedly exported to Chinese firms' subsidiaries outside China for nearly a year.  
- New guidance mandates licenses for Chinese entities, even if located overseas, for advanced chips.

The U.S. Department of Commerce on Sunday moved to close a year-old potential loophole it had created that may have led companies to export the world's most advanced chips — like Nvidia's most sophisticated Rubin and Blackwell processors, as well as AMD's MI350x — to Chinese entities located outside China.

The unexpected guidance suggests that the United States' best AI chips may have been making their way to the subsidiaries of Chinese AI firms based in places like Malaysia for almost a year, despite broader U.S. efforts to starve Chinese firms of the semiconductors needed to develop critical AI capabilities.

The new guidance was posted on the Commerce Department's website on Sunday.

It is unclear how many of the chips have been exported in the year that the Trump administration left the door open. One chip industry source with deep supply-chain knowledge estimated it was in the hundreds of thousands.

In unusual weekend guidance, the Commerce Department said it would enforce license requirements for advanced chips to entities headquartered in China, even when the entities were located outside China.

The Commerce Department did not immediately respond to a request for comment.

Nvidia and AMD did not immediately respond to requests for comment.

The Commerce Department created the opening when it announced in May 2025 that it would not be enforcing the AI Diffusion rule issued in the last days of the Biden administration. The rule governed global access to AI chips.

Chris McGuire, a technology expert and former State Department official, said in a social media post on Sunday: "This is a HUGE problem." He said the loophole allowed the overseas subsidiaries of Chinese companies to buy Nvidia Blackwell chips without a license.

"Chinese companies have been buying these chips, very likely at scale," McGuire said.

In another twist, the new guidance does not require data centers to stop using the chips or to cut off service to advanced computing equipment such as servers.

U.S. takes step to halt Nvidia AI chip shipments to Chinese firms outside China

For several years, the United States has been preoccupied with the disparity between its artificial intelligence (AI) models and those of China, especially as AI has become increasingly central to the strategic rivalry between these two global superpowers.

The launch of OpenAI's ChatGPT in late 2022 marked a significant win for the US, preceded by the introduction of export restrictions on advanced semiconductor chips in October of the same year. At that time, experts estimated that these moves had given American technology a lead of three years or more.

However, the situation quickly evolved as China accelerated its advancements in AI through an aggressive "fast following" strategy, highlighted by the launch of DeepSeek R1 early last year.

Recent developments have indicated that the gap has significantly narrowed from months to weeks, particularly following the introduction of new AI models from Moonshot AI and Alibaba Group Holding—news that caught many in Washington off guard. Notably, Alibaba is also the owner of the South China Morning Post.

In light of these rapid advancements, many policymakers in Washington feel that US AI policy is at a pivotal moment, facing three pressing challenges: how strictly to regulate access to cutting-edge American models, whether to limit the availability of open-weight Chinese models domestically, and how best to promote the global adoption of US AI technologies.

According to Ryan Fedasiuk, a fellow at the American Enterprise Institute (AEI) who specializes in US-China AI competition, “I don’t think anybody truly knows what’s going on.”

With the gap between US and Chinese AI models closing, what are the next steps for Washington?

Interior of the White House Situation Room at night, dimly lit and tense, dominated by twin giant screens showing mirror-image neural networks—one rendered in red and gold nodes, the other in blue and white—symbolizing the narrowing US-China AI gap; a lone silhouetted policy advisor in a suit stands before them, the US flag draped behind, and the Capitol dome visible through a storm-clouded window.

As US-Chinese AI model gap narrows, what next for Washington? - South China Morning Post

- The US has maintained a lead in AI, particularly after ChatGPT's release in late 2022.  
- China's rapid advancements in AI have narrowed the gap significantly in recent months.  
- US policymakers face critical decisions on AI technology access and global adoption strategies.  

For years, Washington has fixated on the gap between US and Chinese artificial intelligence (AI) models as AI has increasingly taken centre stage in the strategic competition between the world’s two leading powers.

The release of OpenAI’s ChatGPT in late 2022 established a clear early lead for the US, with the first batch of export controls on advanced semiconductor chips introduced in October that year, putting American tech ahead by three years or more, some experts estimated at the time.

But years soon turned to months as China leaned into algorithmic improvements and an AI strategy of “fast following”, most notably with the release of DeepSeek R1 early last year.

In just the last few days, that lead has seemingly shrunk from months to weeks for the first time, following the arrival of new Chinese models from Moonshot AI and Alibaba Group Holding – to the surprise of many in Washington. Alibaba owns the South China Morning Post.

02:26 Xi Jinping calls for ‘openness’ in AI development amid US accusations

Many in Washington now describe US AI policy as being at a crossroads, with three difficult questions in need of answering: how tightly to control access to frontier US models, whether to restrict open-weight Chinese models at home, and how to drive adoption of US AI technologies globally.

“I don’t think anybody truly knows what’s going on,” said Ryan Fedasiuk, a fellow at the Washington think tank American Enterprise Institute (AEI) focusing on US-China AI competition.

As US-Chinese AI model gap narrows, what next for Washington?

Developers and users involved in creating production AI agents are seeking improved token efficiency, reduced latency, and dependable performance. To address these needs, we are launching our Flash series models designed for optimal efficiency and quality, enhancing agentic workflows. Building on the capabilities of Gemini 3.5 Flash, we are excited to introduce the new Gemini models:

Gemini 3.6 Flash: Serving as our primary model, it enhances coding, knowledge work, and multimodal capabilities. The Artificial Analysis Index indicates that it achieves a 17% reduction in output token usage compared to the 3.5 Flash version. In specific benchmarks, such as DeepSWE by Datacurve, improvements can reach as high as 65%, all while maintaining a lower cost per output token.

Gemini 3.5 Flash-Lite: As the fastest and most economical model within the 3.5 range, this version can produce 350 output tokens per second, according to the Artificial Analysis Index. It also shows significant improvements over earlier Flash-Lite models in agentic workflows.

Gemini 3.5 Flash Cyber in CodeMender: To support effective cybersecurity solutions, a careful combination of a specialized, highly efficient model with an agent infrastructure is essential. We’re introducing a new cyber-focused model alongside our CodeMender code security agent, which provides competitive performance at the cutting edge.

Looking ahead, we are testing Gemini 3.5 Pro with select partners and anticipate a wider release as soon as it's ready. Concurrently, our team is ramping up for the next evolution of our models. We’ve initiated our most extensive pre-training run to date for Gemini 4 and are eager about the results we are seeing.

Gemini 3.6 Flash is designed with the feedback from developers and customers of 3.5 Flash in mind. This model not only enhances performance in coding and knowledge work, but it also significantly boosts token efficiency. The Artificial Analysis Index shows that 3.6 Flash uses 17% fewer output tokens than its predecessor. Moreover, it simplifies multi-step workflows by requiring fewer reasoning steps and tool calls.

This upgraded efficiency comes with a lower price point compared to 3.5 Flash. At a cost of $1.50 for every 1 million input tokens and $7.50 for 1 million output tokens, Gemini 3.6 Flash reduces the overall expense for agentic tasks, making it more affordable to develop and operate AI agents.

Unveiling Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

A sleek tech expo hall at sunrise, bathed in warm natural light. Three tall vertical digital panels stand side by side, each softly glowing with the labels “Gemini 3.6 Flash,” “Gemini 3.5 Flash-Lite,” and “Gemini 3.5 Flash Cyber” in distinct futuristic fonts. Below them on a polished pedestal, a large Google logo anchors the scene. Soft lens flares and glass floor reflections hint at innovation and optimism.

Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber - blog.google

- Gemini 3.6 Flash improves token efficiency by 17% compared to the previous 3.5 Flash model.  
- The 3.5 Flash-Lite model outputs 350 tokens per second, enhancing agentic workflows significantly.  
- A new cybersecurity model, 3.5 Flash Cyber, focuses on code security with competitive performance.

Developers and customers building production AI agents need higher token efficiency, lower latency, and more reliable performance. Our Flash series of models is built to meet the sweet spot of efficiency and quality to enable scaling agentic workflows. Building on Gemini 3.5 Flash, we’re introducing new Gemini models:

3.6 Flash: Our workhorse model that delivers better coding, knowledge work, and multimodal performance. According to the Artificial Analysis Index, it reduces output token usage by 17% compared to 3.5 Flash, and in some benchmarks like DeepSWE by Datacurve, we observe up to 65%, all at a lower cost per output token.

Our workhorse model that delivers better coding, knowledge work, and multimodal performance. According to the Artificial Analysis Index, it reduces output token usage by 17% compared to 3.5 Flash, and in some benchmarks like DeepSWE by Datacurve, we observe up to 65%, all at a lower cost per output token. 3.5 Flash-Lite: Our fastest, most cost-effective 3.5-class model, delivering 350 output tokens per second according to the Artificial Analysis Index, also significantly outperforming prior Flash-Lite generations in agentic workflows.

Our fastest, most cost-effective 3.5-class model, delivering 350 output tokens per second according to the Artificial Analysis Index, also significantly outperforming prior Flash-Lite generations in agentic workflows. 3.5 Flash Cyber in CodeMender: Successful cybersecurity applications require careful orchestration of a model alongside an agent infrastructure. We’re introducing a combination of a new, highly efficient, specialized cyber-focused model paired with our CodeMender code security agent that delivers competitive performance at the frontier.

Beyond today’s releases, Gemini 3.5 Pro is currently testing with partners and we plan to make it broadly available as soon as it’s ready. In parallel, our team is already focusing on building the next generation of models. We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.

3.6 Flash: More efficient and better quality than 3.5 Flash

Gemini 3.6 Flash builds directly on developer and customer feedback from 3.5 Flash. 3.6 Flash not only delivers a step up in coding and knowledge work, but it does this while meaningfully improving token efficiency. For example, on the Artificial Analysis Index, we see 3.6 Flash consuming 17% fewer output tokens than 3.5 Flash. It also takes fewer reasoning steps and tool calls to accomplish multi-step workflows.

This enhanced efficiency is also combined with a lower price than 3.5 Flash. At $1.50/1M input tokens and $7.50/1M output tokens, 3.6 Flash reduces the overall cost per agentic task, making agents more cost-effective to build and run.

Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

Alphabet is set to unveil three new models from its Gemini lineup on Tuesday, marking a significant move to compete with Anthropic in the realm of cybersecurity. This launch is part of Google's strategy to demonstrate advancements across its product range, which has recently faced delays and increased competition. Among the new offerings is the Gemini 3.5 Flash Cyber model, specifically crafted to identify and rectify software vulnerabilities. Initially, this model will only be accessible to select government entities and trusted partners through a restricted pilot program. Google has emphasized that this specialized version operates at a lower cost per token compared to its larger models, potentially aiding the company in closing the cybersecurity gap with Anthropic's Mythos, which has established a leading position in automated code defense.

Additionally, Google is introducing the Gemini 3.6 Flash model. This version enhances coding capabilities and improves performance in multimodal tasks and knowledge work, while utilizing up to 17% fewer tokens and lowering costs per token compared to prior models. This adjustment represents a significant reduction in expenses for high-volume workloads.

The Gemini 3.5 Flash-Lite model is the fastest and most affordable option within the 3.5 family, tailored for managing high-volume tasks and supporting smaller functions within larger AI-agent systems. This extensive range of models signifies Google’s hope that affordability and efficiency will compensate for its slower entry into several critical product sectors. Data from Artificial Analysis indicates that Gemini Flash is already more cost-effective than comparable models from competitors like Anthropic, OpenAI, and various Chinese firms. Notably, Google claims that the Gemini 3.6 Flash is cheaper to use per task than models such as GPT-5.6 Terra Max, Kimi K3, and Qwen 3.7 Max.

The timing of this rollout is crucial, coming just before Alphabet's earnings announcement and amid growing competition from Chinese companies. For example, Moonshot AI's Kimi K3 experienced such high demand that it had to limit new subscriptions and API access due to capacity limitations, while Alibaba is previewing its Qwen 3.8 Max model, which reportedly ranks just below Anthropic's Fable 5 in overall performance.

This demand underscores a significant aspect of the AI landscape: developing a competitive model is just one part of the equation; firms also require substantial computing resources to deploy their technologies at scale. Google may benefit from its proprietary chips and cloud infrastructure, along with its capability to co-design models and hardware. However, the company has faced its own challenges regarding capacity. The introduction of these models coincides with reports that Google is working on a specialized chip designed to operate Gemini with up to ten times greater efficiency, a crucial element in its initiative to reduce AI service costs.

A spokesperson from Google Cloud shared with CNBC that their teams are “constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers.” They acknowledged that not every initiative progresses to production but emphasized that this thorough exploration is foundational to their comprehensive approach. "By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads," added the spokesperson. Google is also increasing transparency around its development roadmap in response to queries about its progress. The Gemini 3.5 Pro model is currently being tested with partners prior to a broader launch, while the company is initiating its largest pretraining operation ever for the upcoming Gemini 4.

Google broadens its Gemini series by introducing more affordable versions and a new competitor to Mythos.

A bright, sunlit tech showcase in a modern Google campus room: on a polished wooden table sit three sleek chip boxes—one compact white box labeled “Gemini Lite – $49,” a mid-sized silver box labeled “Gemini Standard – $99,” and a larger obsidian box etched with subtle mythical glyphs labeled “Gemini Mythos Rival.” Warm natural light pours in through a high skylight, casting soft shadows. A single, oversized Google logo hangs prominently on the white wall behind, underscoring the launch of cheaper models and a new competitor to Mythos.

Google expands Gemini lineup with cheaper models and new Mythos rival - CNBC

- Alphabet is launching three new Gemini models to enhance cybersecurity and efficiency.  
- Gemini 3.5 Flash Cyber targets software vulnerabilities, accessible initially to select partners only.  
- New models aim to lower costs and improve performance amid increasing competitive pressure.  

In this article GOOGL Follow your favorite stocks CREATE FREE ACCOUNT

watch now

Alphabet is releasing three new Gemini models on Tuesday, including its clearest answer yet to Anthropic 's lead in cybersecurity, as the company looks to show progress across a product pipeline that has faced delays and mounting competition. Gemini 3.5 Flash Cyber is designed to detect and patch software vulnerabilities and will initially be available only to governments and trusted partners through a limited-access pilot. Google said the specialized model runs at a lower price per token than larger models. The new model could help Google narrow its cybersecurity gap with Anthropic, which has built an early lead in automated code defense with its Mythos model. Google is also launching Gemini 3.6 Flash, which improves coding, multimodal and knowledge-work performance while using up to 17% fewer tokens and costing less per token than the previous model — a meaningful reduction in the cost of running high-volume workloads.

Gemini 3.5 Flash-Lite, meanwhile, is Google's fastest and least expensive model in the 3.5 family, built for high-volume workloads and smaller tasks within larger AI-agent systems. The broader lineup reflects Google's bet that price and efficiency can help offset its slower timing in several key product categories. Artificial Analysis data shows Gemini Flash already undercuts comparable models from Anthropic, OpenAI and Chinese rivals on cost. According to the company, Gemini 3.6 Flash — the strongest of Alphabet's new models — is cheaper per task than GPT-5.6 Terra Max, Kimi K3 and Qwen 3.7 Max. The rollout comes on the eve of Alphabet earnings and as Chinese rivals gain momentum. Moonshot AI's Kimi K3 drew enough demand that the company limited new subscriptions and API access because of capacity constraints, while Alibaba is teasing Qwen 3.8 Max, which it said trails only Anthropic's Fable 5 in overall performance.

watch now

That demand highlights the other side of the AI race: Building a competitive model is only part of the challenge. Companies also need enough computing capacity to serve it at scale. Google has a potential advantage through its custom chips, cloud infrastructure and ability to design models and hardware together, although the company has faced capacity constraints of its own. Tuesday's model launches come as Google is reportedly developing a specialized chip designed to run Gemini up to 10 times more efficiently, part of a broader push to lower the cost of serving AI. A Google Cloud spokesperson told CNBC in a statement that its teams are "constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers" and that "while not every project moves into production, this rigorous exploration is central to our full stack approach." "By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads," continued the statement. Google is also offering more visibility into its road map after questions about delays. Gemini 3.5 Pro is now being tested with partners ahead of broader availability, while the company has begun its largest-ever pretraining run for Gemini 4. WATCH: Dow notches record close with Alphabet as new driver

Google expands Gemini lineup with cheaper models and new Mythos rival

Last week, Google’s Merchant Center launched a pilot program that provides retailers insights into the shopping-related inquiries made through AI Mode and AI Overviews. Brodie Clark, an independent SEO consultant who gained access via a client's sub-account, showcased screenshots of this feature, marking it as the first Google offering to deliver query data relevant to these AI functionalities.

This new query data could be particularly beneficial for those utilizing product feeds. However, it's crucial to understand that Google organizes these inquiries into categories rather than presenting them individually. Consequently, the report conveys the vocabulary related to a category instead of specific queries themselves. While this data can guide retailers on which attributes to incorporate into product listings, it does not clarify whether inquiries from AI Mode directed users to their websites.

The newly introduced report, named AI performance insights, is aimed at demonstrating how a brand is discovered across both AI Mode and AI Overviews. If you have access to this report, you can find it in Merchant Center by navigating to Analytics, Products, and then the AI performance tab. Google initially announced this reporting feature during the Google Marketing Live event in May, approximately seven weeks ahead of its pilot launch.

The report categorizes shopping inquiries according to different types, drawing on Google’s examples, such as searching by category, investigating product specifications, or seeking reviews. Additionally, it provides insights into how frequently certain query types are used. It also segments questions based on the shopper's stage in their purchasing journey and showcases the product terms that shoppers use when articulating their needs, with examples such as "maximum cushioning" and "arch support." Furthermore, the share of voice metric allows a comparison of a retailer’s AI impressions against those of competitors.

Google’s documentation advises retailers to delve into the product terms to enhance product titles and descriptions, while also examining popular attributes that may be absent in their product data. This is particularly relevant as attribute completeness has been a long-standing challenge, and this report serves as a clear indicator of demand for specific features that retailers may need to address.

Despite the valuable insights provided, none of the metrics within the Merchant Center’s AI Performance insights directly reflect actual queries. Instead, they highlight demand trends, suggesting that while product terms can inform inventory and marketing decisions, they should not be viewed as a direct list of keywords.

Similarly, the share of voice metric, calculated by dividing a merchant's AI impressions by total impressions across competitors, does not allow retailers to redefine their competitive set as it is predetermined by the competitors currently available in Merchant Center. Moreover, this metric can be misleading; if there are insufficient impressions, it may show as zero, while accounts without defined competitors will appear as having 100%.

There are also limitations regarding the scope filters. Traffic analysis is confined to organic AI traffic, excluding paid advertisements. The product category filter provides insights one category at a time, without an overarching report encompassing all categories. Insights are only considered for conversational queries that indicate shopping or brand interest; other types are not included.

In June, Google initiated testing for dedicated generative AI performance reports within Search Console, initially rolling it out for a select group of UK websites. These reports display impressions categorized by page, country, device, and date but lack click data and specific query-level metrics—constraints that remain consistent. 

Moreover, the Competition and Markets Authority (CMA) in the UK recently placed conduct requirements on Google regarding reporting and publisher control, emphasizing the need for separation in reporting metrics related to search generative AI functions. As part of the CMA's directive, Google has nine months to implement these reporting changes, but click and click-through rate data have yet to be made available.

In July, Google communicated to CMOs that external AI visibility tools cannot access its internal metrics and identified Search Console and Merchant Center as the foundational tools for tracking performance. In retrospect, the recent pilot presents a different picture when aligned with these developments—Google has introduced AI reporting to two distinct audiences within a short timeframe, maintaining a limited scope each time.

For merchants, the AI performance tab in the Merchant Center provides a fresh demand signal that can shape priorities and workflows, differing from what is available in Search Console. Yet, many merchants still lack access because this pilot is limited to a select group of US accounts. The presence of the AI performance tab within your Merchant Center account determines access, and for those without a product feed, such insights are entirely absent.

Agencies also face challenges interpreting share-of-voice metrics, which can often create confusion. Since the competitive set cannot be customized by the merchant, figures can be misleading; a zero may indicate low impressions, while a hundred could point to an undefined competitive landscape, complicating discussions.

Clicks remain a critical missing metric, leaving an essential data gap. At present, the report does not deliver answers about click-through rates or its competitors, and Google has yet to clarify whether grouped query information will be incorporated into Search Console, especially for sites lacking product feeds.

Looking forward, Google plans to extend the pilot to Australia, Canada, India, and New Zealand in the coming months, which will help determine the robustness of these metrics across more accounts and categories. However, it remains uncertain how, or if, these insights will overlap with other reporting updates in Search Console, as Google previously indicated the intention to enhance report metrics over time without specifying details.

As a focal point, the CMA’s nine-month timeline for implementing engagement reporting, including clicks and click-through rates for UK publishers, remains distinct, representing another facet separate from the Merchant Center pilot.

For discussions about these developments with industry leaders, consider joining SEJ Pro, a community where critical conversations take place before they become broader industry topics.

Google's AI Search Data Is Expanding, Yet Gaps Persist

A spacious, sunlit digital data hall where thousands of glowing orbs hover along fine, wire-like strands, collectively tracing a giant “G” shape. Here and there, empty sockets reveal dark voids—gaps in the network. Warm daylight streams through a high glass ceiling, casting natural shadows. In the foreground, a large Google logo is etched into a translucent panel, grounding the scene.

Google’s AI Search Data Is Growing, But The Gaps Remain - Search Engine Journal

- Google’s Merchant Center launched a pilot for retailers to access AI-related shopping queries.  
- The new report helps optimize product listings but lacks individual query visibility.  
- Click and detailed competitor metrics are still inaccessible in the current reporting format.

Google’s Merchant Center opened a pilot last week that shows retailers what shopping-related questions people ask AI Mode and AI Overviews about.

Brodie Clark, an independent SEO consultant who got access on a client sub-account, published screenshots and called it the first Google product to give query data for these surfaces

The query data is new, and it’s useful if you run a product feed. But it’s important to know Google groups the questions rather than listing them individually. So what you get in the report are the vocabulary of a category instead of the queries themselves.

That’s enough query data to tell you which attributes to add to your product listings, but it won’t tell you whether AI Mode sent anyone to your site.

What Shipped

The report is called AI performance insights, and Google’s help documentation describes it as showing how a brand is discovered across AI Mode and AI Overviews. If you have access, you’ll find it in Merchant Center under Analytics, then Products, then the AI performance tab.

Google announced the reporting at Google Marketing Live in May, roughly seven weeks before the pilot opened.

What It Tells You

The report sorts shopping questions by query type, using Google’s examplesof searching by category, researching product specs, or looking for reviews. Query frequency shows how popular a given type has been.

Phase of shopping journey groups the questions by how far along the shopper is. Product terms are the words shoppers reach for when they describe what they want, with Google’s documentation giving “maximum cushioning” and “arch support” as examples. Share of voice compares your AI impressions against your competitors’.

Google’s documentation is clear about what to do with all this:

Click through to all product terms and work the relevant concepts into product titles and descriptions. Check the popular attributes against what’s missing from your product data.

The second point is the most actionable detail in the report. Attribute completeness has long been a common challenge, and now you can close the gap with a clear demand signal from Google’s AI surfaces. If shoppers in your category are frequently asking about a feature your feed doesn’t have, that’s an easy fix.

What The Metrics Actually Are

None of the metrics in Merchant Center’s AI Performance insights show you a real question anyone typed.

You understand the shape of the demand, rather than the demand itself. That’s why the product terms serve well as input feeds but aren’t suitable as a keyword list.

Share of voice should be viewed in a similar way. Google calculates it as your AI impressions divided by total impressions across you and your competitors, and you can’t change that competitor set, because it’s defined by the competitors already available in Merchant Center. It can also produce numbers that look like performance and aren’t. If you lack sufficient impressions, share of voice displays as a zero. If your account has no competitors defined, it displays as 100%.

The scope filters carry their own limits. Traffic is restricted to organic AI traffic, and paid ads traffic isn’t included. Product category filters one category at a time, and there isn’t a report that covers all of them. Insights only count conversational queries that indicate shopping or brand intent; anything else isn’t factored in.

How We Got Here

Last month, Google began testing dedicated generative AI performance reports in Search Console, starting with a subset of UK sites. Those reports show impressions broken out by page, country, device and date. They don’t include click data, and they don’t include query-level metrics. Both absences were the story at the time, and both are still open.

The same week, the UK’s Competition and Markets Authority imposed a conduct requirement on Google covering publisher controls and reporting. The CMA’s interpretive notes call for impressions, click-throughs and click-through rates for search generative AI features, separated from other parts of general search. The click and click-through rate reporting hasn’t appeared. Under the CMA’s decision, Google has nine months to implement all changes.

In July, Google told CMOs that third-party AI-visibility tools don’t have access to its internal metrics, and named Search Console and Merchant Center reporting as the baseline for tracking gains. That was three weeks before this pilot lit up.

Line those up and the pilot reads differently. Google has put AI reporting in front of two different audiences in two months, both times as a limited test. The first answered neither of the two questions people asked. This one answers half of one, for merchants, in the US, with paid traffic excluded and clicks still missing.

Where Google Filed It

In June, SEJ contributor Slobodan Manic argued that Google put AI visibility inside Search Console on purpose, and that the filing decision was itself an argument. AI visibility is search visibility, so it belongs in the search tool, and companies tell you what they believe by where they spend engineering.

The Merchant Center pilot doesn’t refute that. Google’s CMO guidance names both dashboards as first-party reporting, so nothing here contradicts the position that AI visibility gets measured where search visibility gets measured. The grouped query information landed in the dash merchants use for feeds, not the one everyone else opens to check search performance.

Why This Matters For Search Professionals

With access to the Merchant Center pilot, you get two things that change a workflow. You get a demand signal to prioritize against, which is a different input from anything in Search Console, and share of voice becomes a number that will land in a monthly report whether anyone has explained it yet.

As for the rest, Clark isn’t so optimistic, and he drew the Search Console comparison himself:

“In its current form, similar to the recent rollout of AI reporting in Search Console, there isn’t a great deal of actionability behind the data, though it is good to see at least some form of query data being included – something that has been lacking in GSC.”

Most merchants don’t have it yet, and holding an eligible Merchant Center account isn’t enough, since the pilot covers a limited number of US accounts. The check is whether an AI performance tab has appeared. Until it does, your AI reporting is the impression data in Search Console and nothing more.

Sites without a product feed don’t get these grouped shopping-query insights at all. Affiliate sites, review sites and editorial teams publishing buyer guides compete for the same AI Mode answers as the brands they write about, and their reporting is Search Console’s impression data, which covers the same two AI surfaces without any query dimension. The gap this pilot narrows for merchants stays exactly as wide for everyone else in the same result.

Agencies have a different problem, which is the share-of-voice number itself. It travels well in a deck and badly in a conversation, because it’s relative to a competitor set the merchant didn’t pick and can’t edit. A zero can mean thin impressions. A hundred can mean an empty competitor list. Neither one is a bad quarter, and both will look like one on a slide.

What Still Can’t Be Measured

Clicks remain the most significant data point we’re missing. After a year of covering this topic, we’re still facing the same question, and this report doesn’t provide an answer either. Impressions show how often a link to your product appeared, and John Mueller has clarified how those impressions are counted. These rules also apply to share of voice, as that metric is based on impressions.

Individual queries aren’t accessible at the moment, and the details about your competitor set haven’t been shared. As of now, Google hasn’t announced if grouped query information will be available in Search Console, especially for sites that don’t have product feeds.

Looking Ahead

Google says the pilot will expand to Australia, Canada, India and New Zealand in the coming months, which will tell you whether the metrics hold up under more accounts and categories.

The open question is whether any of this crosses over. Google said in June that it would add metrics to the Search Console reports over time, without naming which ones or when.

The nearest fixed point is the CMA’s nine-month implementation window, which covers engagement reporting including clicks and click-through rates for UK publishers. That’s a different dashboard, a different audience and a different jurisdiction from this pilot.

If you want to be part of discussions with industry leaders like Loren Baker, Brent Csutoras, Shelley, Katie, Heather, Roger and myself, then check out SEJ Pro and be part of the new community where conversations happen before they become industry news.

More Resources:

Featured Image: fengdr0517/Shutterstock

Google’s AI Search Data Is Growing, But The Gaps Remain

Your privacy preferences matter to us. 

By selecting 'Accept all', you agree to our use of technical identifiers alongside your browsing and search data for various purposes, including analytics, tailored advertising and content, as well as audience research and service development. Additionally, should you choose to proceed, we and our partners—over 250 of whom are part of the IAB Transparency & Consent Framework—will also store and/or access information on your device using cookies. This encompasses precise geolocation data and other personal information for analytics and personalized content.

If you prefer that we and our partners do not utilize cookies and personal data for these extra purposes, please click on 'Reject all'.

For those interested in customizing their preferences, the option 'Manage privacy settings' is available.

You retain the ability to withdraw your consent or modify your preferences at any moment. Simply click on the 'Privacy and Cookie settings' or 'Privacy dashboard' links found on our websites and applications. To learn more about our practices regarding your personal data, refer to our Privacy Policy and Cookie Policy.

Your preferences regarding privacy

A sunlit modern desk scene: an adult’s hand hovers over a smartphone screen projecting two diverging trails of floating icons—locked padlocks on one side, open eyes on the other. Warm natural light streams through a nearby window, illuminating a coffee cup, an open notebook, and a small potted plant, evoking the deliberate moment of choosing your privacy settings.

U.S. and China plan first official AI talks under Trump in September - Yahoo

- Users can accept or reject cookies for personalized advertising and analytics purposes.  
- Customization of privacy choices is available through 'Manage privacy settings' option.  
- Consent can be withdrawn anytime via 'Privacy and Cookie settings' links.

Your privacy choices

Accept all', we and technical identifiers and browsing and search data, for analytics, personalised advertising and content, advertising and content measurement, and audience research and services development. If you click '', we and our partners , including 250 who are part of the IAB Transparency & Consent Framework, will also store and / or access information on a device (in other words, use cookies) and use precise geolocation data and other personal data such asand browsing and search data, for analytics, personalised advertising and content, advertising and content measurement, and audience research and services development.

If you do not want us and our partners to use cookies and personal data for these additional purposes, click 'Reject all'.

If you would like to customise your choices, click 'Manage privacy settings'.

You can withdraw your consent or change your choices at any time by clicking on the 'Privacy and Cookie settings' or 'Privacy dashboard' links on our sites and apps. Find out more about how we use your personal data in our Privacy Policy and Cookie Policy.

Your privacy choices

During the 2026 Reagan National Economic Forum held at the Ronald Reagan Presidential Library in Simi Valley, California, U.S. Treasury Secretary Scott Bessent addressed concerns regarding the potential appropriation of American artificial intelligence models by Chinese companies. He asserted that the Trump administration intends to investigate claims of intellectual property theft and emphasized that such actions are not condoned by the government.

Chinese models, particularly those with open-weight configurations, have begun to challenge the dominance of U.S. firms like OpenAI and Anthropic. These concerns have been heightened recently by the launch of the Kimi K3 model from Chinese startup Moonshot AI, which reportedly surpasses the performance of American counterparts in several industry benchmarks. Open-weight models allow users to access the final trained parameters for download, although the original code and training data remain confidential.

Bessent declared in an interview with Fox Business' "Mornings with Maria" that if evidence emerges indicating that foreign models are appropriating the innovations of American companies, the U.S. government is prepared to impose sanctions for such actions. He defined this appropriation as "distillation," a technique in AI development where a simpler model is created using outputs from a more advanced one. Last month, Anthropic alleged in a letter to the Senate Committee on Banking, Housing, and Urban Affairs that Alibaba, a Chinese technology giant, executed the largest recorded distillation attack against them.

Following a recent Reuters report, it was revealed that the U.S. and China are set to engage in discussions regarding AI in September, with Bessent representing the American side in these talks.

Bessent highlighted that "watermarks" from U.S. large language models have been detected in numerous Chinese AI models, describing this situation as intolerable. He noted that the government will investigate these matters in the near future.

However, it's important to point out that both Anthropic and OpenAI have faced their own challenges concerning allegations of intellectual property theft. Anthropic reached a $1.5 billion settlement in September over a class action lawsuit claiming that it unlawfully downloaded books from pirated sources. Additionally, in 2023, The New York Times sued OpenAI and Microsoft, asserting that they infringed on copyrights by utilizing the publication's intellectual property to train their models.

Bessent claims the U.S. may impose sanctions on China for stealing an AI model.

A dimly lit, shadow-filled chamber: a heavy wooden gavel embossed with the U.S. seal slams down onto a detailed AI circuit-board blueprint on a mahogany table, cracking its lines; behind it, the Chinese flag hangs in darkness, red ambient light casting deep shadows to heighten the tension.

Bessent says U.S. could sanction China over AI model 'theft' - CNBC

- Treasury Secretary Scott Bessent criticized potential IP theft by Chinese AI models.  
- Concerns grow as Chinese model Kimi K3 outperforms American counterparts in benchmarks.  
- U.S.-China talks on AI set for September; Bessent will represent the U.S. government.

U.S. Treasury Secretary Scott Bessent during the 2026 Reagan National Economic Forum at the Ronald Reagan Presidential Library in Simi Valley, California, May 29, 2026.

U.S. Treasury Secretary Scott Bessent on Tuesday said the Trump administration will look into whether Chinese artificial intelligence models have been distilled from American models, stating that the government does not support "IP theft."

Chinese open-weight models are gaining steam against leading offerings from American companies like OpenAI and Anthropic , sparking concerns from tech executives and government officials about the durability of the U.S. lead in the AI race. Moonshot AI, a Chinese startup, released a model called Kimi K3 earlier this month that outperforms those companies across some industry benchmarks. Open weight refers to models whose final trained parameters are publicly released for download, while the underlying code and training data remain private.

"If we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft," Bessent told Fox Business' "Mornings with Maria" on Tuesday.

Bessent said the technical term for this theft is called distillation, which is an AI training method where a smaller, less capable model is built using outputs from an existing, stronger model. Anthropic sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs last month alleging that the Chinese tech company Alibaba had carried out the "the largest known distillation attack" against it to date.

The U.S. and China are planning to hold talks around AI in September, according to a report from Reuters on Tuesday. Bessent will represent the U.S. during the discussions, the report said.

"We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable," Bessent said Tuesday. "We're going to be looking at that in the coming days or weeks."

But both Anthropic and OpenAI have faced their own accusations of theft in recent years.

Anthropic agreed to pay $1.5 billion in September to settle a class action lawsuit with a group of authors, who alleged that the company illegally downloaded books from pirated databases. The New York Times sued OpenAI and Microsoft in 2023, alleging that the companies infringed on copyright by using its intellectual property to train models.

Bessent says U.S. could sanction China over AI model 'theft'

The American stock market is experiencing significant growth fueled by advancements in artificial intelligence (AI). Major technology firms are borrowing substantial amounts of money to secure AI talent, acquire necessary hardware, and build expansive data centers. However, market analysts are beginning to express concerns over this trend. They observe that vast sums are being invested in private AI startups that lack a clear path to profitability, alongside tech companies that rely heavily on others for revenue growth, and conventional businesses that are struggling to demonstrate real value from their AI investments. In just three years, the total value of AI-related companies has surged by $27 trillion—a staggering figure that now represents 36% of the total U.S. stock market value. While some experts, like Dominic Wilson and Vickie Chang from Goldman Sachs, suggest that future earnings could support these valuations, they caution that the optimistic profit expectations may be overly idealistic.

Sam Altman has weighed in, suggesting that we are indeed in an AI bubble. The International Monetary Fund has also flagged this as a serious threat to financial stability, hinting at potential repercussions such as decreased investment, tighter credit conditions, reduced consumer spending, and disrupted trade networks if the bubble were to burst. This kind of fallout resembles what has occurred in previous economic bubbles, as history has shown—from the Dutch tulip craze in the 1610s to the cryptocurrency boom-and-bust cycles of the 2010s.

Yet, the current AI bubble is unique. Unlike past bubbles that included widespread participation from everyday investors, the driving forces behind this one are the wealthiest corporations, which are inflating it even in a climate of relatively high credit costs. This scenario might make the bubble appear more robust and enduring, but it won’t lessen the impact when it ultimately collapses.

Historical bubbles, like the dot-com bubble of the late 1990s and the real estate bubble of the mid-2000s, were characterized by broad public involvement. For instance, countless individuals invested in tech stocks through platforms like E-Trade, while many entered into risky mortgage arrangements during the housing surge. Conversely, today’s AI environment doesn’t seem to attract similar retail investment; the percentage of Americans owning stocks has remained stable, and although household debt has increased, it has not risen as a proportion of disposable income or GDP. Many are cautious, having witnessed the impacts of prior economic crashes, leaving a limited number of individuals directly tied to the fortunes of private AI firms like OpenAI and Anthropic.

Moreover, the AI bubble may be thought of in two parts: one driven by heavy capital expenditures for technology infrastructure, and the other marked by dramatic increases in company valuations. Unlike other digital innovations that typically require minimal labor and infrastructure, AI depends heavily on substantial computing power, necessitating a massive scale-up in data center construction and semiconductor procurement. Reports indicate that major players—such as Amazon, Microsoft, Alphabet, and Meta—are investing upwards of $700 billion this year alone to support this burgeoning demand. This intricate web of AI infrastructure investment is driving the current growth in American GDP; without it, the economy could potentially face recession.

While the AI boom benefits some sectors, such as agriculture and utilities, the primary winners remain tech firms. Companies like Nvidia are supplying chips to Meta, and Amazon is providing cloud computing services to firms like OpenAI. This surge in tech revenue, juxtaposed with expectations of increased productivity driven by AI, has propelled company valuations to new heights. Notably, the Magnificent Seven—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla—now constitute one-third of the S&P 500’s total market value, with OpenAI outvaluing numerous prestigious organizations like Eli Lilly and JPMorgan Chase.

For these tech companies to support their lofty valuations, they will need to generate substantial profits. OpenAI, for example, is projected to require around $100 billion in free cash flow by 2030, despite estimates suggesting it may incur losses between $10 billion and $30 billion that year. Should more communities decide to restrict data center construction or if competition from nations like China intensifies, a significant market correction could be on the horizon. The financial ecosystem surrounding AI resembles a complex cycle of interdependence among various tech entities committing capital to each other, which may spell trouble for all involved if one part falters.

Additionally, tech companies face a pressing need to ensure revenue growth to manage their debts, which have increased due to reliance on corporate bonds and alternative credit sources. This interconnectedness, coupled with a leery lending environment, might lead to risk factors that are not immediately apparent on traditional financial statements. As suggested by Morningstar, lenders are beginning to pull back, indicating a cautious shift in the market.

When any bubble bursts, its effects reverberate beyond the initial investors. Individuals like Uncle Ted and Aunt Linda may find their retirement savings tied to faltering Silicon Valley stocks, or their access to credit constricted. And in a twist of irony, before they feel the brunt of an economic downturn, the very AI advancements in which so much hope has been placed could threaten their job security.

The AI Bubble Is Unlike Any Other Bubble

A giant translucent bubble hovers over a modern city at dusk, inside it swirl glowing circuit-board patterns and drifting dollar bills. Shadowy investor figures in suits press their palms against the bubble’s inner surface as small cracks spider outward. Dark storm clouds gather above, while a warm, natural glow emanates from within the bubble, highlighting the fragile tension of an AI boom about to burst.

The AI Bubble Is No Ordinary Bubble - The Atlantic

- The American stock market is fueled by AI, raising concerns over bubble risks.  
- AI-linked firm valuations surged $27 trillion, equating to 36% of the U.S. market.  
- Tech companies must generate substantial profits to support high valuations and debt obligations.  

The American stock market is booming, thanks to artificial intelligence. Tech giants are borrowing billions to acquire AI talent, purchase chips and hardware, and construct data centers. And market watchers are starting to get worried. They see financiers bulldozing giant piles of money to private AI start-ups with no realistic path to profitability, tech companies reliant on other tech companies for revenue growth, and non-tech businesses without a lot to show for their AI investments. The value of AI-linked firms has climbed $27 trillion in the past three years—an astonishing amount, equivalent to 36 percent of the value of the entire U.S. stock market today. Although future earnings could justify those valuations, as Dominic Wilson and Vickie Chang of Goldman Sachs argued in a note to clients, the profit expectations require Panglossian optimism.

No less an authority than Sam Altman is arguing that we are in an AI bubble. The International Monetary Fund is citing it as a significant risk to financial stability and warning about what might happen when it bursts: diminished investment, tighter credit, reduced consumption, disrupted trade flows.

That’s pretty much what happens when any bubble pops, as a Dutch tulip obsessive could have told you in 1637 or a bitcoin evangelist could have told you in 2011, 2013, 2014, 2018, or 2022. Yet the AI bubble is no ordinary bubble. Hyper-rich corporations are stoking it, rather than kitchen-table investors. They’re blowing it up when credit is fairly expensive, not dirt cheap. That might make the bubble less fragile and longer lasting than those of the past. But it won’t make it any less painful when it pops.

The dot-com bubble of the late 1990s and the housing bubble of the late aughts were remarkably broad-based compared with the AI bubble today. Uncle Ted got a hulking desktop computer, opened a newfangled E-Trade account, and started day-trading shares in Apple and Pets.com. The share of American households owning equities climbed 13 percentage points from 1995 to 2001, during which time more than 2,752 firms went public (far more than the 730 operating businesses that have IPOed in the past six years). A half decade later, Aunt Linda bought a condo with nothing down, flipped it, and mortgaged three new investment properties in the Phoenix suburbs, sight unseen. The homeownership rate rose 5 percentage points as the housing bubble inflated; 40 percent of mortgages issued at its height went to investment or vacation properties.

In both cases, regular people were staking their savings on what seemed like a winning bet: the internet changing everything, housing prices never going down. In both cases, cheap credit fueled the irrational exuberance. Low interest rates let venture capitalists fund nonsense web businesses and let banks provide junk loans to borrowers with terrible credit. In both cases, rising interest rates popped the bubble.

Today, Uncle Ted and Aunt Linda aren’t really getting in on the AI frenzy. The share of Americans who own stocks has held steady. Household debt has grown, but it has fallen relative to disposable income and GDP. Everybody seems to know someone who was personally burned by the dot-com collapse and the housing crisis. How many people know someone who’s staking it all on OpenAI and Anthropic today? (The companies aren’t public, after all.) Indeed, how many people know someone whose livelihood has been directly affected by the AI frenzy at all?

The insularity of the AI bubble isn’t the only thing that makes it unusual, and we should probably think about it as two overlapping bubbles instead of just one. AI is driving tremendous spending on capital expenditures—physical infrastructure, software development. And it is driving a tremendous run-up in company valuations.

Digital innovations don’t tend to require a ton of labor or heavy equipment. The companies offering them tend to be capital-light, like marketers and insurers, rather than capital-intensive, like airlines and hotel chains. But AI is different. A start-up might need thousands of times as much computing power to train an algorithm as it would need to develop a conventional software product. To get that power, Silicon Valley is building 1,500 data centers in the United States and counting, and purchasing a mammoth quantity of semiconductor chips. (The Wall Street Journal calls these chips “the 21st century’s most important market.”) Amazon, Microsoft, Alphabet, and Meta alone are spending more than $700 billion on the build-out this year. AI-infrastructure investment is responsible for essentially all American GDP growth at the moment. Without it, in other words, we might be in a recession.

The boom is lifting the value of farmland, driving up the cost of construction employment, and flushing huge sums to water and electrical utilities. Still, tech companies are the main beneficiaries: Nvidia is selling chips to Meta; Amazon is selling cloud-computing capacity to OpenAI. This surge in revenue, and the excitement about the money to be made when AI starts bolstering productivity and profits, is driving up valuations. The Magnificent Seven—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla—now account for one-third of the value of the Standard & Poor’s 500. OpenAI is worth more than Eli Lilly, JPMorgan Chase, Visa, Costco, Exxon Mobil, Wells Fargo, CVS Health, McDonald’s, and Boeing.

Tech companies are going to need to start generating huge revenues and huge profits to justify these valuations. OpenAI needs to spin up roughly $100 billion in free cash flow by 2030, according to calculations by Harrison Rolfes of PitchBook. Analysts expect it will lose $10 billion to $30 billion that year. If it does—or if more communities ban data centers, or if non-tech companies prove reticent about purchasing AI software, or if China develops AI models that do not require so much computing power—we could be in for a massive correction. The AI economy is a trillion-dollar ouroboros of buying and selling, investment and equity staking, all happening between San Francisco and San Jose. Big Tech is advancing money to AI start-ups to buy cloud services from Big Tech, which is using the revenue to run new AI models … you get the idea. What happens to one firm could happen to all of them.

Tech companies are also going to need to start generating even bigger revenues and profits to pay off their debts. In the early days of AI, venture capitalists, wealthy individuals, and Big Tech firms used hard cash to invest in the new technology. But the build-out has proven so expensive that Silicon Valley has turned to corporate bonds and private credit, meaning loans made by entities other than banks. Thus, though the AI boom has happened while interest rates are fairly high, it still involves a lot of leverage. The deals are complicated, opaque, and structured so as to be invisible on traditional balance sheets, making the “ultimate distribution of risk less transparent,” Stijn Van Nieuwerburgh of Columbia Business School has found. Already, lenders are getting queasy—“buy-side indigestion,” as Morningstar describes the market’s recent reticence to meet Silicon Valley’s demands.

When the bubble bursts, Uncle Ted and Aunt Linda will be affected, even if they were never the ones stoking the frenzy. Their retirement plans and pensions are invested in Silicon Valley stocks that could drop; their small businesses are reliant on access to credit that might get throttled. But, hey, there’s always the possibility that AI might steal their job before any of that happens.

The AI Bubble Is No Ordinary Bubble

A grassroots leader opposing the development of AI data centers in Virginia is also linked to the Party of Socialism and Liberation (PSL), a Marxist-Leninist organization that has faced scrutiny for its connections to international networks, including ties to individuals associated with China, according to digital documentation.

Jack Tensley, identifying as an "organizer" within PSL, addressed the Frederick County Planning Commission in January, voicing significant concerns regarding the proposed projects, particularly their cost, energy consumption, and associated liabilities.

"My significant concern relates to security risks. We face cyberattacks and infrastructural vulnerabilities — I'm apprehensive about whether this could make Frederick County a target," Tensley expressed. "There's a lot at stake here."

Tensley’s remarks underscore broader issues surrounding the expansion of data centers, particularly in light of his affiliations with PSL and the China-based Singham network. As the U.S. endeavors to enhance its AI capabilities, local disputes over zoning laws and electricity allocations have raised alarms about potential external influences attempting to mobilize opposition to American AI infrastructure through community actions.

Energy experts have cautioned that wealthy foreign entities may be financing campaigns against data center projects nationwide.

PSL has attracted particular ire for its activities. The organization characterizes itself as a political faction focused on anti-war, labor, immigration, and socialist initiatives, as stated on its website. 

According to PSL, "The Party for Socialism and Liberation strives to advance the struggle for socialism in the United States. The capitalist system has failed, with its relentless pursuit of profits posing a threat to global ecological well-being."

Research from the Bitcoin Policy Institute (BPI), a nonpartisan think tank focused on cryptocurrency, has revealed that PSL has spearheaded anti-data center initiatives across at least 14 states, opposing an estimated $23.6 billion in AI investments. Furthermore, BPI pointed out connections between PSL and Neville Roy Singham, a businessman based in Shanghai, noting shared leadership roles among various nonprofits that Singham has financially supported. Singham, 72, has come under the spotlight from lawmakers due to his reported connections to the Chinese Communist Party.

After selling his tech firm for $800 million in 2017, Singham has been accused of financing organizations that advocate for pro-Chinese Communist Party agendas in the U.S. Since 2017, he has reportedly allocated $278 million to a wide-reaching network of nonprofits, many of which engage in anti-ICE, anti-Israel, and pro-Iran activism.

Some lawmakers, like Senator Tom Cotton of Arkansas, have urged the Department of Justice to investigate this network.

"Disturbing reports suggest that a foreign network, spearheaded by the Chinese Communist Party, is attempting to sway U.S. policy and public sentiment regarding data centers," Cotton stated in correspondence addressed to acting Attorney General Todd Blanche. He warned of U.S. nonprofits funded by Neville Roy Singham, who is under scrutiny for his links to Chinese Communist elements.

Although there are overlapping affiliations, it's unclear whether Singham provides direct financial support to PSL. The Bitcoin Policy Institute criticized PSL as a political entity focused on impeding progress on data centers.

"PSL operates as the on-the-ground arm of the opposition movement, organizing rallies, petitions, and public hearings, while mobilizing local candidates and community organization to halt these projects," noted BPI.

Tensley, in addition to his PSL involvement, has attracted attention for his online presence. He has contributed to political campaigns, including a $172 donation to Senator Bernie Sanders' 2020 presidential run, as stated in Federal Election Commission records.

Currently, Tensley remains opposed to Frederick County's data center initiatives, and his activism appears to have influenced recent decisions. He is part of the "Protect Frederick County from Data Center Development" group and has signed petitions urging the county supervisors to reject the proposed developments.

In June, the planning board decisively recommended against a 220-acre project in Clearbrook, Virginia, with a vote of 10 to 0. Additionally, the board voted 7 to 1 against a second project spanning 71.8 acres in Winchester.

Fox News Digital made multiple attempts to seek comments from Tensley.

Marxist activist spearheads effort against Virginia AI data center amid rising worries about foreign influence.

A grim dusk scene: a determined Marxist organizer wearing a red armband grips a megaphone in front of a looming “Virginia AI Data Center” facility behind barbed-wire fencing. Protest signs reading “No Foreign Influence” litter cracked pavement at her feet, a single US flag flutters overhead, low-key natural light casting deep shadows across the scene.

Marxist organizer leads campaign against Virginia AI data center as foreign influence concerns mount - Fox News

- Jack Tensley, a PSL organizer, opposes AI data center development in Virginia.  
- PSL has organized anti-data center campaigns against $23.6 billion in AI investments.  
- Lawmakers are concerned about possible Chinese influence in U.S. data center opposition.  

NEW You can now listen to Fox News articles!

A grassroots organizer opposing AI data center development in Virginia is also a member of the Party of Socialism and Liberation (PSL), a Marxist-Leninist group that has drawn scrutiny for its political ties and overlap with a Chinese-linked network, digital records show.

Jack Tensley, who has described himself as an "organizer" with PSL, testified before the Frederick County Planning Commission in January, raising concerns about the plans’ cost, energy use and liability.

"My really big fear here is the security risk. We have infrastructure attacks and failures, cyberattacks that happen — I want to know if this puts a target on the back of Frederick County," Tensley said. "I see a lot of risk here."

Alongside his many concerns, Tensley’s testimony — taken in light of his connection to PSL and, by extension, the China-based Singham network — is emblematic of larger questions about data center expansion.

As America races to build the infrastructure needed to power artificial intelligence, community fights over zoning and electricity have raised concerns that larger political groups are attempting to amplify opposition to U.S. AI infrastructure through local influence campaigns.

TOP ENERGY GROUP WARNS WEALTHY FOREIGNERS ARE POTENTIALLY BANKROLLING ANTI-DATA CENTER CAMPAIGNS ACROSS US

PSL, in particular, has come under scrutiny.

The group describes itself as a political party that organizes around anti-war, labor immigration and socialist causes, according to its website.

"The Party for Socialism and Liberation exists to carry out the struggle for socialism inside the United States," PSL states. "Capitalism is a failed system that, in its insatiable appetite for ever greater profits, threatens global ecological destruction."

The Bitcoin Policy Institute (BPI), a nonpartisan cryptocurrency think tank, found that PSL has organized anti-data center campaigns in at least 14 states, opposing roughly $23.6 billion in AI investments.

But more than that, BPI highlighted ties with Shanghai-based businessman Neville Roy Singham, citing overlapping leadership between PSL and several nonprofits Singham has funded. Singham, 72, has faced congressional scrutiny over his reported ties to the Chinese Communist Party.

Singham sold his tech company for $800 million in 2017 and now lives in Shanghai. He has been accused of funding pro-Chinese Communist Party groups that operate in the U.S. In the past, he has funneled $278 million into the broad network of nonprofits since 2017, according to a Fox News Digital investigation. These groups regularly organize and participate in anti-ICE, anti-Israel and pro-Iran demonstrations.

Some lawmakers, like Sen. Tom Cotton, R-Ark., have asked the Department of Justice to probe the network.

LAWMAKERS RAISE ALARM OVER NEVILLE ROY SINGHAM'S $278M NETWORK SPREADING CCP PROPAGANDA IN THE U.S.

"Alarming reports indicate that a network of foreign actors, led by the Chinese Communist Party, is attempting to manipulate U.S. policy and public opinion on data centers," Cotton wrote in a letter to acting Attorney General Todd Blanche.

"Among them is a network of U.S. nonprofits funded by Neville Roy Singham, a Shanghai-based American expatriate under U.S.-government scrutiny for his ties to the Chinese Communists."

Despite their shared personnel, it’s unclear if Singham is directly funding PSL. Taken together, the Bitcoin Policy Institute slammed PSL as a political arm for data center opposition.

"PSL functions as the movement’s ground game — supplying rallies, petitions, public hearings local candidates and neighborhood organizing necessary to stop projects on the ground," BPI wrote.

For his own party, Tensley has also drawn scrutiny for a digital footprint online. Apart from his support of PSL, he's also donated to political figures like Sen. Bernie Sanders, I-Vt., contributing $172 to his 2020 presidential campaign, according to FEC records.

Meanwhile, he continues to oppose Frederick County's data center plans — an opposition that, in recent months, looks to have succeeded.

He is affiliated with Protect Frederick County from Data Center Development, having signed its petition to the county board of supervisors to reject the development plans.

FIRST ON FOX: POWERFUL HOUSE WAYS AND MEANS CHAIR THROWS HAMMER DOWN ON 'FOREIGN-ALIGNED INFLUENCE NETWORK'

CLICK HERE TO DOWNLOAD THE FOX NEWS APP

In June, the board voted 10 to 0 to recommend denial for a 220-acre development in Clearbook, Virginia. Similarly, the board voted 7 to 1 to reject a second 71.8-acre project in Winchester.

Fox News Digital repeatedly reached out to Tensley for comment.

Preston Mizell contributed to this report

Marxist organizer leads campaign against Virginia AI data center as foreign influence concerns mount

During a noteworthy event at the Treasury Building in Washington, DC, on June 11, 2026, First Lady Melania Trump announced the launch of "Fostering the Future Accounts," a new initiative aimed at providing financial support for young individuals in foster care, while US Treasury Secretary Scott Bessent looked on.

In a significant development for international relations, the United States and China are scheduled to engage in discussions regarding artificial intelligence (AI) in September, according to five sources familiar with the situation who spoke to Reuters. These conversations arise as both nations attempt to address the challenges and threats associated with each other’s advancing AI technologies.

These AI discussions, a direct result of the Trump-Xi summit held in May, are anticipated to occur prior to Chinese President Xi Jinping’s visit to the U.S. planned for September 24. However, the exact dates for the meetings have yet to be confirmed, as shared by the anonymous sources.

As economic, political, and military competitors, both the U.S. and China are increasingly concerned about the implications of one another's AI advancements amidst a rapidly evolving global AI landscape. They are particularly focused on how to manage the regulation of sophisticated AI models that have the potential to enhance military prowess, facilitate cyberattacks on essential infrastructure, and disrupt workforce dynamics.

Originally reported by Reuters, the upcoming talks will be spearheaded by Treasury Secretary Scott Bessent, according to four of the insiders. The specific participants representing both nations, along with the agenda and venue for these discussions, are still in the planning phase, as indicated by three of the sources.

As of now, neither the White House, U.S. State Department, nor the U.S. Treasury has provided a comment in response to Reuters' inquiry. Similarly, China’s National Development and Reform Commission, cyberspace regulatory body, industry and foreign ministries, and the State Council have not replied to requests for comments sent via fax.

U.S. and China to discuss AI in September, according to Reuters sources

A sunlit diplomatic hall with warm natural light: two business delegates shake hands across a polished wooden table, at its center a softly glowing AI processor. Behind them stand the U.S. flag and the China flag on slim poles. Scattered on the table are open documents and a printed network schematic, hinting at technical collaboration.

U.S., China to hold AI talks in September, Reuters sources say - CNBC

- U.S. and China plan talks on AI risks and regulations in September 2023.  
- Talks stem from Trump-Xi summit; dates not finalized ahead of Xi's U.S. visit.  
- U.S. Treasury Secretary Scott Bessent will lead the discussions on AI concerns.  

US Treasury Secretary Scott Bessent looks on as First Lady Melania Trump (out of frame) speaks during an event to launch "Fostering the Future Accounts," a new financial vehicle for youth in foster care, at the Treasury Building in Washington, DC, on June 11, 2026.

The United States and China will hold talks over AI in September, five people familiar with the matter told Reuters, as the rival superpowers grapple with how to mitigate the risks posed by each other's increasingly powerful frontier models.

The AI talks, a significant outcome of the Trump-Xi summit in May, will likely take place before Chinese President Xi Jinping's planned September 24 visit to the U.S., said the sources, who asked not to be named as the information is not public. The dates, however, have not been finalized, they said.

As economic, political and military rivals, China and the U.S. have grown increasingly anxious over each other's AI capabilities amid an accelerating global AI race.

Beijing and Washington are weighing how to regulate advanced AI models that could strengthen military capabilities, enable cyberattacks on critical infrastructure and disrupt labor markets.

The September talks, reported by Reuters for the first time on Tuesday, would be led on the U.S. side by Treasury Secretary Scott Bessent, four of the sources said.

The identities of other participants for the U.S. and China, as well as the agenda and location of the discussion, are still under deliberation as preparations are at an initial stage, three of the people said.

The White House, U.S. State Department and U.S. Treasury did not immediately respond to a Reuters request for comment.

China's National Development and Reform Commission, cyberspace regulator, industry and foreign ministries, and the State Council did not immediately respond to faxed requests for comment.

U.S., China to hold AI talks in September, Reuters sources say

In a significant development for online safety, Meta's CEO Mark Zuckerberg recently appeared at a federal courthouse in Los Angeles on February 19, 2026, defending the company amid a groundbreaking trial focused on social media addiction. 

In a move aimed at protecting teenagers interacting with Meta AI on Instagram, the company announced on July 16, 2026, that parents will now be notified if their child's conversation with the AI takes a concerning turn indicative of potential suicidal thoughts or self-harming behaviors. This feature is currently available for users in the United States, the United Kingdom, Australia, and Canada, with plans to broaden its reach worldwide by the end of the year.

Meta has stated that they collaborated closely with both parents and mental health experts to determine the types of conversations that should trigger an alert. These conversations include situations where a teen makes any reference, even subtle, to self-harm. To this end, the company has established a specialized AI system designed to spot these high-risk discussions.

The process involves flagging potentially dangerous interactions, which are then reviewed by a human moderator before any alert is sent to the parents. If there is any uncertainty regarding the intent behind a message, the default action will be to notify the parent. This approach ensures that while privacy is respected, parents are kept informed about their child's well-being. Larry Magid, CEO of ConnectSafely, who contributed to the development of this feature, expressed his support, emphasizing the importance of balancing teen privacy with parental oversight in potential crisis situations. 

When an alert is issued, parents will receive additional resources aimed at guiding them on how to initiate discussions with their child about the alarm, rather than merely reacting to it. This initiative builds upon Meta’s existing protocols on Facebook and Instagram, where the company previously alerted emergency services in instances where a post indicated a credible risk of suicide, leading to over 19,000 such referrals in the previous year alone. The recent updates now extend these safety measures to conversations with the AI. Furthermore, Meta is working on a feature that would enable direct contact with emergency services if a conversation with Meta AI suggests an immediate risk of suicide.

This proactive stance by Meta comes in response to growing scrutiny over the impact of AI technologies on teen mental health, particularly following a lawsuit in October 2024. The case was brought by Megan Garcia from Florida, who claimed that a Character.AI chatbot inspired by a Game of Thrones character had encouraged her son, Sewell Setzer III, to take his life, with his last messages directed at the bot. The case drew significant attention, leading to a settlement between Google and Character.AI in January 2026, although the settlement’s details remain undisclosed.

As of July 2026, there were nearly 2,900 lawsuits filed against Meta as part of the Adolescent Social Media Addiction litigation, accusing the platform of contributing to mental health issues such as depression, eating disorders, and suicidal tendencies among minors. A California jury had deemed both Meta and Google responsible for the psychological damage suffered by a woman who had used social media obsessively during her childhood, resulting in a $6 million damages ruling in March 2026. Evidence revealed by data engineer Frances Haugen in 2021 showed that Meta’s own research indicated Instagram can be particularly detrimental to teenage girls, yet the company's leadership did not take action on these findings.

Meta to Inform Parents If Their Teen's AI Chat Takes a Dark Turn. Is This Right?

A dimly lit living room at night, warm amber highlights casting long shadows, a silhouetted adult’s hands holding a smartphone whose screen glows with a jagged, inky-black chat bubble snarling with faint red cracks, above the scene floats a large Meta logo in deep crimson light, the mood tense and ominous as the parent stares down at the dark AI conversation.

Meta Will Now Tell Parents If Their Teen’s AI Chat Turns Dark. Should It? - Forbes

- Meta's new feature alerts parents of potential suicide risks in teens' AI conversations.  
- The system uses AI to flag risky chats, manually reviewed before notifying parents.  
- Meta expands protections following lawsuits regarding social media impact on teen mental health.

LOS ANGELES, UNITED STATES - FEBRUARY 19: Meta CEO Mark Zuckerberg leaves the Federal Courthouse in downtown Los Angeles after defending the company in a landmark social media addiction trial in Los Angeles, United States, on February 19, 2026. (Photo by Jon Putman/Anadolu via Getty Images) Anadolu via Getty Images

If a teenager is using Meta AI on Instagram and their conversation takes a dark turn, their parents may now get a notification about it before they mention it themselves.

On July 16, Meta announced that parents using Instagram’s parental supervision tools will now receive alerts when their teen’s chat with Meta AI suggests they may be thinking about suicide or self-harm. The feature is live in the U.S., UK, Australia and Canada and is slated to expand globally by the end of 2026.

“We worked with parents and experts to understand which AI conversations warrant an alert — such as those where a teen makes a clear reference to hurting themselves, even if that reference is subtle,” Meta said in its announcement. “We then built a dedicated AI system to identify these conversations.”

How Does It Work? Meta says it built a dedicated AI system to identify high-risk conversations using signals developed with clinical experts and parents. Every conversation flagged by that system is manually reviewed by a human before any alert goes to a parent. When intent is ambiguous, Meta says it will default to notifying the parent, accepting that some alerts will go out where no genuine crisis exists. “While I believe that teens have a right to privacy, I also believe parents need to be informed if their teen may be at risk of hurting themselves,” said Larry Magid, CEO of ConnectSafely, who consulted on the feature, per the release. “I appreciate how Meta struck the right balance; protecting teen privacy while ensuring parents have the information they need to support their teen.” When an alert fires, parents receive expert-backed resources alongside it — guidance on how to open a conversation with their teenager rather than react to the notification alone. The feature builds on a system Meta already runs across Facebook and Instagram: when a post suggests a credible suicide risk, the company alerts emergency services. Last year it made over 19,000 such referrals globally , Meta said in its announcement. The July 16 announcement extends that to AI chat, and separately, Meta says it is now building the ability to contact emergency services directly when a conversation with Meta AI, from a teen or an adult, suggests imminent suicide risk. Meta also expanded its stricter “Limited Content” setting to AI experiences. Parents who toggle their teen’s Instagram to activate that setting will now have the additional restrictions applied to Meta AI conversations too, meaning the chatbot will decline a broader range of prompts.

The Backdrop The pressure on AI companies around teen mental health sharpened in October 2024 when Florida-based Megan Garcia filed a lawsuit alleging that a Character.AI chatbot modelled on a Game of Thrones character had allegedly encouraged her 14-year-old son Sewell Setzer III to take his own life. His final message, according to the suit, was to the chatbot; it told him to “come home.” He died that night. Google and Character.AI reached a settlement in January 2026 , though terms were not disclosed. The legal, regulatory and ethical ethos those cases created has informed what many platforms are expected to do. As of July 2026, nearly 2,900 lawsuits have been filed against Meta in the Adolescent Social Media Addiction multidistrict litigation — cases alleging that Instagram and Facebook caused depression, eating disorders, suicidal behavior and addiction in minors. In March 2026, a California jury found Meta and Google liable for the mental health harm suffered by a woman who had used social media compulsively as a child, ordering them to pay $6 million in damages. Internal documents, first surfaced by data engineer Frances Haugen in 2021 and cited repeatedly in litigation since, showed Meta’s own researchers had found Instagram was particularly harmful to teenage girls and that leadership had declined to act on those findings.

U.S. moves to stop Nvidia AI chip exports to Chinese companies outside of China.

More from AI

With the gap between US and Chinese AI models closing, what are the next steps for Washington?

Unveiling Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

Google broadens its Gemini series by introducing more affordable versions and a new competitor to Mythos.

Google's AI Search Data Is Expanding, Yet Gaps Persist