A US restriction on foreign access to Anthropic’s latest AI models has become more than a fight over one company’s products. It is now a visible test of how governments, AI labs and foreign allies respond when a frontier model is described as powerful enough to deserve special caution.
The ban covers Anthropic’s newest models, Mythos and Fable. Critics argue that Anthropic’s own public warnings about advanced AI helped create the conditions for that decision, while others see the move as evidence that regulation of the most capable models is arriving faster than many in Silicon Valley expected.
Why Anthropic’s warnings matter now
According to FT analysis, Anthropic has spoken about the risks of advanced AI much more often than OpenAI this year. The research examined official statements, social media posts and articles written by Anthropic or its chief Dario Amodei, then compared them with communications from OpenAI and Sam Altman.
The difference was stark. Five in every 1,000 words used by Anthropic in 2026 related to risk, regulation or restrictions. For OpenAI and Sam Altman, the equivalent figure was 0.6 words per 1,000, eight times lower.
That gap has become politically sensitive because Washington last week barred foreign nationals from using Mythos and Fable. Some technologists have blamed the restriction on Anthropic’s repeated emphasis on AI’s risks to society, especially around Mythos.
Yann LeCun, Meta’s former chief AI scientist and one of AI’s pioneers, described Amodei’s approach as “ridiculous fear-mongering” and wrote, “One reaps what one sows.” The criticism captures a central tension for AI companies: warning about dangerous capabilities can support calls for safety, but it can also invite government limits on who may use the technology.
What the FT analysis found
The FT built lists of terms including “harmful,” “dangerous” and “misaligned,” then calculated how often those terms appeared in communications by each company or its CEO. It also used sentiment analysis to compare the positive and negative tone of the messages.
The word-level findings show how different the two companies’ public language has been in 2026:
- In Anthropic communications, “risk” appeared 336 times.
- “Safeguard” appeared 121 times.
- “Vulnerability” appeared 128 times.
- At OpenAI, those same terms appeared 30, 33 and 10 times respectively.
Anthropic’s communications were still largely positive overall, according to the analysis. But they were less positive than OpenAI’s. The FT also found that Anthropic has softened its language significantly since 2023, with its use of risk and regulation-related language roughly halved from the same period in 2023.
That context matters because the current argument is not simply that Anthropic talks about risk. It is that Anthropic talks about risk more often than a major rival, while also developing models that officials and critics now treat as sensitive.
Mythos put safety claims under pressure
Mythos is central to the dispute. Anthropic has described it as capable of discovering critical cyber security gaps and initially limited access on safety grounds to certain US organisations. The company had also been working with government officials on a controlled rollout before releasing Mythos more widely earlier this month.
Days before the export ban, Dario Amodei published a long blog post on his personal website arguing that regulators were moving too slowly on AI. He wrote that “In the last few months… the evidence of AI’s incredible power, as well as its risks, has become undeniable,” and said Mythos demonstrated “very real risks to cyber security, creating the potential for disruption of the financial sector, critical infrastructure and national security.”
The FT analysis said that post was not the most negative Anthropic publication it reviewed. A news item from April 2025 about rare AI behaviours, including sabotage or providing information about weapons, contained roughly three times more negative language.
Still, Mythos attracted unusual attention. News coverage of the model, announced in April, was significantly higher than coverage of other models released this year, according to data from AlphaSense. Media mentions surged after Mythos was unveiled and rose again this week following the export ban.
The dispute is bigger than one company
The ban has unsettled parts of Europe and Silicon Valley. Executives and officials fear the Trump administration may be willing to restrict non-US access to frontier models. That possibility makes the Anthropic case an early signal of how the US may oversee increasingly powerful AI systems.
Some industry figures have also criticized Anthropic’s handling of government talks. David Sacks, former AI tsar to the US government, wrote on X that a “credible trusted partner had approached the administration with a way to circumvent the guardrails placed on Fable. He claimed Anthropic downplayed the concerns, forcing the government to “reluctantly” impose the ban.
The relationship between Anthropic and senior government figures was already strained. The export ban follows public clashes over issues including the use of Anthropic’s technology in domestic surveillance and lethal autonomous weapons. In February, the Pentagon named Anthropic as a supply-chain risk to national security, and the two sides are in litigation over that designation.
Anthropic declined to comment.
Regulation, innovation and allied access
The government’s move may align with parts of US public opinion. YouGov polling cited in the source showed that the majority surveyed agreed that effective regulation was important even if it slowed technological advances.
Outside the US, the issue is being read through the lens of cooperation among allies. French President Emmanuel Macron said the Anthropic dispute had “clarified the stakes” for the US and its allies in the G7. He called for “stronger regulation of artificial intelligence” and warned against the risk of “non-cooperation among democracies.”
Lennart Heim, an independent AI policy researcher who formerly worked at think-tank Rand, said the US government’s response did “not inspire confidence.” He pointed to the tension between an administration that has positioned itself as pro-innovation and a decision to ban foreign access to what he called the most advanced US model.
That contradiction is why the Anthropic export ban now matters beyond Mythos and Fable. It raises a practical question for every frontier AI developer: if a company publicly stresses danger to make the case for careful deployment, it may also strengthen the case for government control over who gets access.