Anthropic Increases Support for Public First Action to $40 Million

Anthropic has contributed an additional $20 million to Public First Action, bringing its total financial support to $40 million. This funding is dedicated exclusively to public education and policy missions aimed at establishing sensible AI safeguards and cannot be used to influence elections for federal, state, or local office.

Accelerating AI Capabilities and Associated Risks

Rapid improvements in AI model capabilities are creating urgent security vulnerabilities that necessitate immediate government action. Anthropic cites the example of Claude Mythos Preview, which discovered thousands of high-severity software vulnerabilities across every major operating system and browser. To mitigate the risk of these vulnerabilities being exploited, Anthropic released the model to a limited set of cyber defenders via Project Glasswing.

While advanced models offer significant potential benefits—including accelerated drug discovery, disease treatment, scientific progress, extended human lifespans, and economic growth—Anthropic asserts that protections against catastrophic risks must be established first to realize these gains. The company emphasizes that governments require immediate time to build the capacity to manage these risks while capturing the benefits.

Proposed Policy Frameworks for Frontier AI

Anthropic argues that transparency alone is insufficient given the pace of AI advancement. The company advocates for a comprehensive approach to risk mitigation as detailed in the Advanced AI Framework, which includes the following requirements:

• Verification and Enforcement: Governments should have the ability to verify safety claims, enforce safe practices through civil penalties, and block the deployment of models that pose a serious risk of catastrophic harm.
• Testing and Evaluation: Frontier AI developers must test models for catastrophic risk, maintain transparency regarding findings, submit models to independent evaluation, and operate robust security programs.

National Security and Global AI Leadership

To maintain the leadership of the United States and its democratic allies, Anthropic supports specific national security measures to prevent advanced technologies from benefiting authoritarian regimes. These proposed measures include:

• Export Controls: Tightening controls on advanced chips and semiconductor manufacturing equipment.
• Access Restrictions: Curbing illicit model access and preventing distillation attacks.

Anthropic views these policy frameworks as a starting point and uses its support of Public First Action to increase the salience of the urgent policy debate surrounding AI safety and national security.