The White House and Anthropic are negotiating new AI security rules to evaluate vulnerabilities in advanced models. These discussions represent a shift toward technical standards after recent export controls disrupted the company’s operations.
According to administration officials, the proposed framework will assess the severity of security flaws. Specifically, this system will guide potential government intervention when vulnerabilities are detected in new software.
The negotiations follow export controls that forced Anthropic to suspend access to its Fable 5 and Mythos 5 models. Notably, these restrictions were triggered by a perceived security flaw known as a jailbreak.
Establishing AI security rules for Models
The administration is working to establish clear AI security rules to manage risks that could affect national security. Meanwhile, Anthropic CEO Dario Amodei disagreed with officials over the severity of the initial jailbreak vulnerability.
Consequently, both parties are developing a common set of benchmarks to evaluate future security incidents. These benchmarks will measure how safeguards were bypassed and the practical consequences of any breach.
Export Controls and Model Suspensions
On Friday, talks had effectively collapsed after Anthropic rejected demands to de-deploy Fable. The company argued 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. Consequently, this action forced the company to pull the affected models from the market temporarily.
Rebuilding Communication Channels
Over the weekend, Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross held discussions with Anthropic co-founder Tom Brown. These lengthy calls initiated a week of in-person meetings in Washington.
Furthermore, Anthropic sent senior researchers and safeguards experts to the Commerce Department on Monday. These experts are working directly with officials to resolve the outstanding technical disputes.
Future Outlook for Industry Standards
Other technology companies and G7 leaders recently discussed similar risk measurement frameworks in France. Therefore, these negotiations could influence how governments regulate artificial intelligence systems globally.
While the export controls remain in place, the ongoing talks suggest a collaborative path forward. Both sides continue to refine the technical benchmarks to ensure stable cybersecurity practices.
Sarah Heck, head of public policy, and Tom Brown are leading the discussions for the company. Their goal is to establish a predictable framework under the new AI security rules for future apps and software deployments.
Neither Anthropic nor the White House immediately responded to requests for comment regarding the progress of the negotiations. However, the shift toward technical standards indicates that both sides are now making progress.
