Anthropic is donating another $20 million to Public First Action
Key Points
- Additional $20M to Public First Action (total $40M)
- Claude Mythos found thousands of high‑severity vulnerabilities
- Calls for enforceable transparency, independent evaluation, export controls
Summary
Anthropic is contributing an additional $20 million to Public First Action (bringing total support to $40M) to fund nonpartisan public education and policy work on AI safety. Funds are restricted from being used to influence elections. Anthropic cites rapid advances in model capabilities — including Claude Mythos Preview uncovering thousands of high‑severity software vulnerabilities — as the immediate rationale for accelerating public policy engagement and funding independent education and advocacy.
Key Points
- Donation purpose: support Public First Action’s public education and policy mission (not for electoral activity).
- Recent technical trigger: Claude Mythos Preview discovered thousands of high‑severity vulnerabilities; Anthropic limited release to vetted cyber defenders via Project Glasswing.
- Policy position: transparency alone is insufficient; Anthropic advocates for its Advanced AI Framework requiring testing for catastrophic risk, independent evaluation, public transparency of results, and enforceable civil penalties.
- National security and governance: supports tighter export controls on advanced chips and measures to curb illicit model access and distillation to prevent authoritarian misuse.
- Practical implications for engineers:
- Build and document robust security programs for model development and deployment.
- Integrate independent evaluation and red‑team testing into release pipelines.
- Prioritize threat modeling for critical systems (healthcare, energy, infrastructure) and compliance with emerging transparency and export control rules.
- What’s next: Anthropic’s donation is intended to raise policy salience and accelerate government capacity to capture AI benefits while containing risks.