AI security – The White House is working with Anthropic on a framework to assess how severe security flaws in new AI models are—aimed at guiding potential government intervention. The talks come after export controls forced Anthropic to suspend access to its latest models,
When Anthropic hit a breaking point with the U.S. government over a jailbreak vulnerability, the argument quickly stopped being technical and started becoming structural: who decides whether a flaw is severe enough to shut a model out of the market?
That standoff is now shifting into something more measurable. The White House and Anthropic are working on a framework that would assess the severity of security flaws in new AI models and guide potential government intervention. a senior White House official and an administration official familiar with the matter said on the condition of anonymity to discuss it with POLITICO.
The effort comes after the White House imposed export controls on Anthropic—controls that forced the company to suspend access for all users to Fable 5 and Mythos 5, its latest powerful AI models, over a perceived security flaw that the industry calls a jailbreak.
Inside the dispute, severity was the fault line. Administration officials and Anthropic CEO Dario Amodei disagreed over the severity of the jailbreak, POLITICO previously reported. The disagreement was complicated by a plain reality: the technology has outpaced the government’s infrastructure for defining and assessing disputes like these.
The negotiations point to a broader push to establish guardrails for newer and more powerful models. Some fear that if vulnerabilities go unchecked, the stakes could reach economic and national security. Administration officials’ push for standardized evaluation is also grounded in a shared understanding that no AI model can be completely immune to hacking—an idea that Anthropic initially echoed when defending its model. and one that other leading AI companies and country leaders also relayed at G7 meetings earlier this week in France.
In practical terms. talks between the White House and Anthropic—led on the company side by Sarah Heck. head of public policy. and Tom Brown. cofounder—are aimed at developing common benchmarks. Those benchmarks are expected to cover future jailbreaks. including the extent to which safeguards were bypassed. the capabilities exposed. and the practical consequences of the breach.
Anthropic and the White House did not immediately respond to a request for comment.
Even as those discussions move toward standards, the export controls on Anthropic have not yet been lifted.
The pressure built after talks effectively collapsed on Friday. Anthropic rejected demands to de-deploy Fable, arguing the vulnerability was limited and did not amount to a meaningful security flaw. The White House responded by imposing export controls that barred foreign users from accessing the model. forcing the company to pull it from the market.
Over the weekend, the tone changed. Senior administration officials and Anthropic leaders held a series of lengthy calls with Anthropic cofounder Tom Brown. Commerce Secretary Howard Lutnick. and National Cyber Director Sean Cairncross. Those conversations led to nearly a week of in-person meetings in Washington.
Anthropic sent senior researchers and safeguards experts to the Commerce Department on Monday to patch things up with administration officials.
For now. the negotiations show how quickly the conversation is moving—from arguing about the meaning of a jailbreak to trying to turn it into a repeatable measurement that a regulator can apply. The story also lays bare the friction at the center of the relationship: the government is trying to build a ruleset fast enough to keep up. while companies are pushing for standards that reflect what they view as real-world risk rather than worst-case assumptions.
White House Anthropic AI security export controls Fable 5 Mythos 5 jailbreak Commerce Department Howard Lutnick Sean Cairncross Dario Amodei Tom Brown Sarah Heck G7 national security