Anthropic has recently taken down its two latest AI models, Fable 5 and Mythos 5, in response to an export control order issued by the Trump administration. This development has ignited significant discussions regarding AI policy and the concept of digital sovereignty.
In the latest episode of the TechCrunch Equity podcast, hosts Sean O’Kane, Rebecca Bellan, and I delved into the reasons behind the administration's actions against Anthropic and the potential repercussions for the wider AI industry.
Sean observed that Anthropic seems to have had a particularly strained relationship with the Trump administration compared to other prominent AI firms, suggesting that its competitors might not face the same jeopardy.
However, Rebecca highlighted that a notable group of cybersecurity specialists has penned an open letter urging the administration to rescind the order, arguing that the decision compromises crucial cybersecurity capabilities for U.S. network defenders.
We speculated whether this situation might ultimately play in Anthropic's favor, especially considering Rebecca's remark that "everyone loves a bad boy."
Here are highlights from our conversation, streamlined for clarity.
Rebecca Bellan: As many listeners might be aware, the U.S. government mandated Anthropic to take its two newest models offline—Fable 5, which was more broadly available, and Mythos 5, which was accessible to existing Mythos users. The administration issued a letter last Friday that referenced "national security concerns," but without any elaboration on those concerns. Anthropic was instructed to prevent any foreign nationals from using these models, leading the company to decide to withdraw them altogether, as they couldn't definitively identify foreign national status, especially since many of their employees are foreigners.
Reports suggest that the administration's attention to this issue was sparked by Amazon researchers who allegedly discovered a way to bypass the restrictions on Fable 5. After Amazon CEO Andy Jassy raised these concerns, the situation escalated quickly.
Sean O’Kane: The timeline was incredibly rapid, especially transitioning from Friday into the weekend, coinciding with the administration's negotiations related to its military actions in Iran.
Rebecca: It certainly felt like a distraction on a Friday evening in New York.
Sean: To provide some broader context, Anthropic has seem to have consistently clashed with the Trump administration, which does set them apart from other leading AI labs. Do you think this incident will affect those other companies?
Anthony Ha: It's worth noting that both media reports and analyses from independent security experts indicate the security risks associated with Anthropic aren’t particularly unique. Much of this seems to stem from a poor relationship between Anthropic and the Trump administration. While this situation might benefit other companies — allowing them to operate without drawing the administration's ire — it ultimately creates a precarious regulatory environment.
Rebecca: This situation feels somewhat vindictive. After the government categorized Anthropic as a supply chain risk, ongoing litigation seems to have made the White House eager for any justification to go after the company. Many cybersecurity experts contend that the circumstances should never have warranted an export control order. They've collectively urged Trump to reverse the decision, emphasizing that removing advanced cybersecurity tools from U.S. defenders is indeed perilous. Anthropic has argued that similar vulnerabilities exist in other AI models as well.
Cynically, one might wonder if this pause on Anthropic is merely an opportunity for competitors to catch up.
At the same time, there are opinions that argue Anthropic had it coming, as they previously indicated that their models are too dangerous for general release while still promoting their own powerful offerings. Just a week before Fable’s introduction, they were calling for a slowdown in AI development due to potential dangers.
Anthony: This entire scenario epitomizes the larger discourse surrounding AI. Figures like Sam Altman and Jensen Huang are seeking to alleviate tensions, questioning why they have been vilified. However, their messaging over the past couple of years—claiming they've crafted an "all-powerful" AI—doesn't exactly help their cause.
Notably, Anthropic's discussion around Mythos portrays it as an exceedingly potent model, one that poses too great a danger to be made openly available. If we are to take this seriously, it invites heightened scrutiny.
It’s apparent that Anthropic is dissatisfied with the situation, and while we shouldn't overestimate the potential benefits for them, there have been instances where previous conflicts with the administration ended up favorably for the company. After a past controversy, interest in their Claude model surged, as users began to view it as a more ethical alternative to ChatGPT.
In a similar vein, Anthropic may find its models perceived as even more formidable in light of this latest incident.
Rebecca: Absolutely. The notion that "we are so dangerous" can generate intrigue. Everyone wants a taste of that forbidden fruit, especially when the former president is involved.
