U.S. Clampdown Silences Top AI Overnight

Washington just forced one of America’s top AI companies to shut down its newest model worldwide, raising real questions about security, secrecy, and who controls the future of this powerful technology.

Story Snapshot

The U.S. government issued an export-control order blocking foreign nationals from Anthropic’s advanced Fable 5 and Mythos 5 AI models on national security grounds.
Anthropic responded by disabling both models for everyone worldwide, not just foreign users, disrupting millions of customers overnight.
Officials fear hackers could “jailbreak” the system and find serious software flaws, but the company says the evidence is narrow and unproven.
The fight highlights a bigger struggle over who will set the rules for frontier AI: elected leaders, the security state, or private tech giants.

Government clamps down on Anthropic’s most powerful AI

Federal officials used export-control powers to order Anthropic to block foreign access to its newest artificial intelligence systems, called Fable 5 and Mythos 5, citing national security concerns about possible software attacks.^[1] The directive hit late on a Friday evening, when many people had already left work, and it focused on foreign nationals who might probe the model for hidden ways to break into computer networks.^[6] Officials say the models’ advanced code skills pushed them to act before a disaster.

Anthropic had just launched Fable 5 and Mythos 5 as its most capable tools, able to run tasks on their own for longer than older models and help write and test complex software.^[3] Experts in Washington and Silicon Valley were already warning that this kind of system could make hacking tools easier to build and use.^[2] The Commerce Department framed the move as a classic export-control case, where the risk of powerful technology in the wrong hands is judged too great to wait for proof of actual harm.

Anthropic shuts models off for everyone, then pushes back

The government order officially targeted foreign nationals, not American citizens, but Anthropic announced it would disable both models for all users worldwide.^[2] The company said it was following a binding legal directive and could not keep the tools online for one group while blocking another without huge technical and legal headaches. In its public statement, Anthropic stressed that it was “suspending access” and “working to restore it as soon as possible,” framing this as a pause, not a confession that the systems were broken beyond repair.

Company leaders also made clear they disagreed with the scope of the clampdown. They argued the government’s concern centered on what they called a narrow and not yet proven “jailbreak” that might, under some conditions, help expose software vulnerabilities.^[1] From Anthropic’s point of view, they had already built guardrails and tracking rules for these high-end models, including data retention on advanced traffic, to catch misuse.^[6] They warned that yanking a widely used commercial product based on early findings sets a worrying precedent for future AI innovation.

Security fears, overreach worries, and what this means for citizens

National security officials say they acted because frontier AI can cut both ways: it can help defend networks, but it can also supercharge attacks if hostile actors learn to exploit it.^[1] They argue export limits on cutting-edge tools are nothing new and compare this to controls placed on advanced chips or strong encryption in past years.^[2] Their message is simple: once a dangerous AI skill spreads overseas, there is no easy way to pull it back or see who is using it against American targets.

Critics, including Anthropic itself, warn that secret threat reports and vague fears can now be used to shut down tools that many peaceful users depend on. They point out that outside experts still cannot review the claimed jailbreak in detail, and that the public is left to choose sides without full information.^[1] For everyday Americans, the deeper question is whether unelected security officials or giant tech companies will set the boundaries of new AI power, and how the Constitution’s limits on government fit into this new fight over code, speech, and control.