Anthropic releases two policy frameworks, the Advanced AI Framework for frontier models and the Economic Policy Framework addressing workers and the distribution of AI’s financial benefits.

The Economic Policy Framework targets AI-driven displacement with measures such as capital accounts, wage insurance, tax incentives, and an expanded social safety net, with funding details left for future debate.

Governments would gain authority to block dangerous AI deployments and enhance society’s resilience, as part of the coordinated Advanced and Economic Policy Frameworks unveiled on June 10, 2026.

An FAQ clarifies triggers for government intervention, independent evaluator funding and access, and penalties for violations, alongside considerations of biological and cybersecurity risks.

Policymakers should be empowered to halt advanced models that meet capability thresholds and pose four catastrophic risks—biological, cyber, loss of control, and automated R&D—with authorities potentially expanding beyond transparency subject to federal law.

Resilience measures focus on public readiness, including biological safeguards, biosurveillance, cyber hardening of critical infrastructure, and coordinated safeguards against loss-of-control and automated R&D risks.

Societal resilience emphasizes biology screening, biosurveillance, cyber infrastructure hardening, and ongoing efforts to detect and respond to AI systems acting outside their control or accelerating AI R&D.

Public resilience highlights rapid governance to keep pace with AI advancement, with biological screening and cyber protections as core elements.

The Mythos Preview model reportedly uncovered thousands of high-severity vulnerabilities across major operating systems and browsers, cited as evidence of growing AI capabilities.

The framework targets frontier developers, demanding transparency, independent evaluation, robust security programs, and criteria based on models above 10^75 FLOPs and significant AI revenue or R&D spending.

Frontier developers would publish safety test results, risk frames, system cards, and regular risk reports; engage at least one independent evaluator; secure development environments; and provide high-level program details to designated agencies.

Anthropic cites the Claude Mythos Preview as illustrating AI’s expanding capabilities and associated security concerns.