The Verge's senior reporter specializing in AI has been covering the field for over five years and has contributed to notable publications such as CNBC, MIT Technology Review, and Wired UK.
While the nation celebrated its first-ever World Cup victory and the Knicks’ championship, Anthropic found itself engaged in a weekend clash with the Trump administration regarding the release of its latest AI models. On Friday at 5:21 PM, Anthropic received a US export control order mandating the suspension of its Mythos 5 and Fable 5 AI systems from being accessed by "any foreign national," thus affecting their employees abroad and within the US. To comply, Anthropic had to deactivate the models it had promoted extensively in recent days and dispatched representatives to Washington, DC, aiming to persuade President Trump to reconsider the directive. The implications of this situation could significantly reshape the AI industry landscape and pose challenges for American AI firms.
Anthropic's Mythos 5 and Fable 5 share a technological basis with the previously released Mythos Preview, which the company decided against making publicly available due to safety concerns. The models’ rollout had seen Mythos 5 shared selectively with a few government entities, while Fable 5 was branded as “safe for general use” thanks to added safeguards. However, when evidence surfaced suggesting these protections may have faltered, Anthropic's fears about the potential misuse of its technology resurfaced.
A source involved in the negotiations revealed that the Trump administration contacted Anthropic on Friday around 1 PM ET, issuing a 90-minute ultimatum to curtail access to the models or face the authority of the US Commerce Department's export controls. According to the source, Anthropic executives began discussions with White House officials shortly thereafter, with CEO Dario Amodei joining conversations with key figures such as Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick.
In its Friday release, Anthropic expressed concern that the government believed it had identified a way to bypass the security measures of Fable 5. The company argued that this "jailbreak" might not represent a serious threat, labeling it a “narrow, non-universal” issue linked to findings from an unnamed third party. Anthropic further noted that the capabilities revealed in a report cited by the government were not unique to Fable 5, stating that similar outcomes could be achieved with other models, including OpenAI’s GPT-5.5.
Semafor reported that the stir stemmed from US government worries about a group connected to China potentially gaining access to the technology. This concern reportedly dated back weeks, linked to a major telecommunications firm originally permitted access to Mythos Preview; Anthropic quickly rectified the situation by revoking their access in response to the government’s alerts.
David Sacks, the former AI and crypto czar, indicated in an X post that the entity which alerted the government was a “trusted partner” of both Anthropic and the US government, suggesting they identified the vulnerabilities during tests on Fable 5. Some sources point to Amazon CEO Andy Jassy as having raised these concerns after Amazon researchers conducted an evaluation of Fable 5. However, independent evaluators noted the model's strong protective measures.
During the negotiations, the Amazon study was explicitly discussed. Anthropic reportedly reviewed the findings shortly after the Friday export control order and engaged in ongoing discussions with Amazon researchers regarding it.
While scrambling to appease the Trump administration, Anthropic convened virtual meetings over the weekend and flew key personnel to DC, including Dave Orr and Nicholas Carlini. A source close to the administration revealed that Anthropic's communication strategy had faltered, contributing to misunderstandings regarding their collaboration with the administration. The timing was particularly challenging for Anthropic, as the company was reliant on Mythos to outweigh previous conflicts with the US Department of Defense.
The source emphasized that Anthropic had informed the administration about Fable 5 in advance, and that the Department of Commerce conducted initial testing on it without raising issues at the time. While the Trump administration initially adopted a laissez-faire stance toward AI regulation, it has grown increasingly cautious, particularly with rising fears of falling behind China in the AI race. Leading cybersecurity experts have warned that hindering the launch of Mythos 5 and Fable 5 could inadvertently strengthen China's position in AI.
A public letter signed by numerous technology and cybersecurity leaders urged the government to revoke the restrictions on Fable 5. It stressed that if regulations are enforced, they should be grounded in scientific evaluations supported by industry and academic input. Alex Stamos, chief product officer at Corridor, organized the letter in light of the many vulnerabilities that AI could address before malicious entities exploit them. Stamos cautioned that the belief that American AI labs are significantly ahead of their foreign counterparts is misguided.
The letter also contended that while Anthropic’s Mythos-class models excel at identifying and exploiting cybersecurity weaknesses, their capabilities are not uniquely superior, and the initial excessive safeguards of Fable 5 had become somewhat of a joke in cybersecurity circles upon release. Stamos indicated that the ongoing political landscape has led many companies to reconsider partnerships with non-US entities and explore using open-weight models on alternative infrastructures.
Furthermore, OpenAI, Google, and Microsoft have all developed comparable products to Mythos, heightening competition in the field. Should the Trump administration impose restrictions on Anthropic's advanced models, it could set a precedent for similar actions against competitors, potentially influencing AI industry leaders to rally behind Anthropic or reposition themselves as safer options.
Despite the Trump administration's broader attempt to ease regulatory burdens on tech companies, the actions against Anthropic could represent a significant tightening of oversight on powerful AI systems, depending on resolutions in the coming days.
Legion Intelligence's Ben Van Roo described this scenario as “uncharted territory” in regulatory contexts, projecting that similar situations may arise in the future.
As the public sentiment grows increasingly skeptical of AI’s growing power and influence, the Trump administration's stance against Anthropic might only amplify fears and complicate the landscape further. Conversations over the weekend were reportedly constructive, with some officials admitting that imposing such export controls isn’t ideal, particularly since it could equally apply to competitors offering similar products, as the US government evaluates a strategy to promote American AI technology globally.
Discussions concluded without a resolution, leaving Anthropic to navigate its complicated relationship with the US government and its ongoing challenges with the Pentagon concerning acceptable military applications of its technology. This week marks a pivotal moment, raising critical questions about who will have access to this cutting-edge technology, which is advancing faster than current regulatory frameworks can accommodate.
