Anthropic says it has taken its latest AI models offline to comply with U.S. directive
Anthropic disabled its Fable 5 and Mythos 5 AI models on Friday following a U.S. government directive to block access for foreign nationals. This action follows a recent executive order by President Donald Trump to vet advanced AI systems for national security risks, specifically targeting models capable of exploiting software vulnerabilities.
Why is the U.S. government restricting AI models like Mythos?
The U.S. government is targeting AI models that possess “dual-use” capabilities—tools that can both create software and destroy it. According to Anthropic, the Mythos model can find software flaws at scale, a capability the company warned could pose serious risks to national security and public safety if misused.
This restriction is the most significant step to date in U.S. efforts to control AI exports. It stems from an executive order signed by President Donald Trump 10 days prior to the Anthropic shutdown. That order established a framework for the federal government to vet the national security risks of advanced AI for up to a month before public release.
How do AI export controls impact global cybersecurity?
Export controls on AI create a fragmented digital landscape where the most powerful tools are restricted by nationality. Tiff Macklem, governor of the Bank of Canada, stated that the ability of AI to both expose and exploit vulnerabilities “puts a premium” on strong cybersecurity protections.
The trend suggests a shift toward “AI sovereignty,” where nations treat model weights and API access as strategic assets similar to nuclear technology or advanced semiconductors. When models like Mythos are pulled offline, it prevents foreign adversaries from using the AI to automate the discovery of zero-day vulnerabilities in critical infrastructure.
However, these restrictions also create a gap for global defenders. If only one nation controls the most advanced “flaw-finding” AI, other countries may struggle to patch their systems before an attacker—using a different, unrestricted model—finds the hole.
What happens when AI developers clash with national security directives?
The friction between private AI labs and government mandates is increasing. Anthropic stated it disagrees with how the U.S. government handled the Friday directive, noting that the administration did not specify the exact national security concerns driving the order.
Anthropic argued that the government should block unsafe deployments through a process that is “transparent, fair, clear, and grounded in technical facts.” The company characterized the current action as a “misunderstanding” and expressed hope to restore access quickly.
This conflict highlights a growing tension: AI companies want global scale and open collaboration, while the U.S. government is moving toward a “vetting” model. While the Trump executive order stated that participation by developers would be voluntary, the directive sent to Anthropic suggests the government may use more direct pressure when national security is cited.
Comparison: Voluntary Framework vs. Direct Action
| Executive Order (10 Days Prior) | Anthropic Directive (Friday) |
|---|---|
| Participation is voluntary. | Models taken offline immediately. |
| Vetting period up to one month. | Immediate restriction of foreign nationals. |
| Focus on framework establishment. | Enforcement of export controls. |
Frequently Asked Questions
Why were Fable 5 and Mythos 5 taken offline?
Anthropic disabled the models to comply with a U.S. government directive aimed at preventing foreign nationals from accessing advanced AI capabilities that could threaten national security.
What makes Mythos 5 a security risk?
According to Anthropic, the model can identify software flaws at scale, which could be used by malicious actors to attack economies or public safety systems.
Is the U.S. government forcing all AI companies to do this?
The recent executive order established a voluntary framework for vetting AI risks, but the specific directive to Anthropic shows the government is actively intervening in the deployment of high-risk models.
What do you think about the balance between AI innovation and national security? Should the government have the power to shut down models overnight? Let us know in the comments below or subscribe to our newsletter for more deep dives into AI policy.