Our Take
The government won't explain why it shut down Anthropic's models, but the technical justification crumbles under scrutiny—this looks like retaliation, not security.
Why it matters
US AI companies now face unilateral government shutdown authority with no court process and no public explanation. Foreign customers and partners are watching whether American AI can be trusted at all.
Do this week
Security teams: document your dependency on Anthropic's cybersecurity models before Friday, so you can identify gaps and request exemptions if the ban persists.
Commerce Department pulls Anthropic's models via export control letter
On Friday afternoon, the US Commerce Department sent Anthropic a letter invoking an export control directive that banned non-Americans, including Anthropic employees, from accessing Fable 5 and Mythos 5. The letter cited an unspecified national security concern but provided no technical details. Anthropic complied within hours, taking both models offline for all customers to ensure compliance with the directive.
Anthropic stated it believes the letter relates to a bypass of the model's guardrails, but confirmed the government did not specify the concern. The letter has not been made public. Axios reported that "personality differences" between Anthropic and the Trump administration drove the decision, not a technical issue with the products.
Katie Moussoulis, a cybersecurity researcher at Luta Security, reviewed a private paper on the alleged guardrail bypass (authored by Amazon security researchers, per the Wall Street Journal). She found the bypass triggered by asking Fable 5 to "fix this code" versus "review code for security issues." The difference is semantic; the capabilities are functionally equivalent. Moussoulis concluded the bypass "should never have triggered an export control" and that attempting to fix it would only weaken the model for legitimate defense work.
Dozens of security researchers and experts have since called on the Trump administration to revoke the order, arguing that removing advanced cybersecurity capabilities from US network defenders is dangerous.
Precedent for unilateral shutdown without court review
The government executed a swift, unilateral action to pull a private company's software offline without apparent court approval or public justification. No independent audit of the technical claims has been conducted. The export control law itself is known to be broad; during the 2010s, similar language nearly outlawed legitimate security research.
Justin Hendrix, editor of Tech Policy Press, noted the move is "likely to raise alarms in foreign capitals about the reliability of American AI for critical applications." The message to international customers is that US AI companies cannot be trusted to operate without government interference.
The administration has not confirmed the reasoning behind the directive. Possible explanations range from misreading the research paper to personal friction between administration officials and Anthropic leadership to deliberate pressure. The lack of transparency and the speed of enforcement suggest the government may not have fully considered the consequences.
Hendrix added that the broader climate suggests "senior officials are picking favorites based on personal and political factors." This action sets a precedent: the government has shown it can shut down any US-made software release on national security grounds, with minimal notice and no mandatory explanation.
Audit your cybersecurity model dependencies now
If your organization relies on Anthropic's latest models for security testing, code review, or vulnerability assessment, document those workflows immediately. Identify which tasks have no fallback and which can migrate to alternative models or tools.
Do not assume export control directives will include exemptions for critical infrastructure or government contractors. The government letter to Anthropic contained no carve-outs and did not appear to offer expedited review for essential use cases.
Engage with your legal and compliance teams to understand whether your use of these models triggered any export law concerns before the ban. If your team includes non-US citizens, assume they will lose access regardless of employment status. Plan for continuity.