Anthropic Proposes Framework for Government Oversight of Frontier AI Models

Anthropic has released two policy proposals to prepare for rapid AI advancement, Anthropic News reports. The published details focus on the Advanced AI Framework, which outlines governance for increasingly capable systems, while the Economic Policy Framework addresses preparing workers and broadly sharing financial benefits.

Designed primarily for the U.S. federal government, the Advanced AI Framework would apply to models trained using more than 10²⁵ floating-point operations (FLOPs), developed by companies earning more than $500 million in AI-related revenue or spending more than $1 billion on AI R&D. It calls for granting the government legal authority to block or deter dangerous deployments of models posing significant catastrophic risks, backed by civil penalties tied to global annual revenue that escalate with repeated violations.

The framework identifies four categories of catastrophic risk: biological weapons development, large-scale cyberattacks on critical infrastructure, loss of control over autonomous systems, and automated AI R&D that amplifies other risks. Anthropic argues that the rapid pace of AI improvement means transparency requirements alone—such as those recently enacted in California and New York—are no longer sufficient.

For frontier developers, the proposal mandates publishing testing summaries, safety frameworks, system cards, and regular risk reports; engaging at least one qualified independent evaluator to review risk assessments; and maintaining robust security programs to protect model weights and training infrastructure against external and internal threats.

The framework also recommends societal resilience measures, including gene synthesis screening and biosurveillance for biological threats; hardening critical software and replacing legacy systems for cyber threats; and further research into detecting and containing AI systems that act outside developer control.

Anthropic urges policymakers to engage on these issues immediately and states that Congress should not preempt state AI laws unless it enacts federal legislation at least as strong as the proposed framework.