White House, Anthropic Set AI Security Rules After Export Controls

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, POLITICO reported. The talks shift the focus from export controls on Fable 5 to managing security risks in the models themselves.

The framework would classify vulnerabilities by severity and determine when the government steps in. The discussions follow a period where export restrictions on Fable 5, a model Anthropic developed, dominated the policy conversation. Now the emphasis is on what happens after a model is built – how to evaluate and respond to the risks it poses.

Anthropic has pushed for clear rules on when the government can intervene, according to people familiar with the talks. The company wants predictability: a set of criteria that trigger a response, not ad hoc decisions after a model is released. The White House has signaled openness to a structured approach but has not committed to specific thresholds.

The shift matters for the broader AI industry. If the framework becomes a template, other developers would face similar evaluation requirements. The rules could set a precedent for how the U.S. government handles AI safety, moving beyond the current patchwork of voluntary commitments and sector-specific guidance.

Export controls on Fable 5 remain in place. The new framework does not replace them. It adds a second layer: security assessment after deployment, not just before it.

A White House official declined to comment on the timeline for finalizing the framework. Anthropic did not respond to a request for comment.