Anthropic Doubles Down on AI Policy Spending With $20M More to Public First Action

Anthropic has pushed its total backing of Public First Action to $40 million, adding a fresh $20 million contribution to a nonpartisan group focused on educating the public about artificial intelligence policy. The move signals that the AI lab views political engagement, not just model releases, as core to its strategy on safety regulation.

Public First Action works across party lines, engaging Republicans, Democrats, and Independents on what it calls AI safeguards policy. Anthropic first backed the organization in February 2026, and this latest injection of funds effectively doubles that initial commitment. The company has framed its support as part of a broader push to see states adopt transparency rules that require AI developers to disclose more about how their systems are built, tested, and deployed. Several states have already passed such laws, according to Anthropic, and the company has publicly backed that legislative trend.

The donation announcement arrived alongside a separate but related disclosure: Anthropic said it released a preview version of a model called Claude Mythos to a limited group of cyber defenders through Project Glasswing. That tool reportedly surfaced thousands of high-severity software vulnerabilities, including flaws found in every major operating system and browser. While the vulnerability-hunting news is technically distinct from the political donation, Anthropic is presenting both moves as part of the same narrative: AI systems are powerful enough to both find catastrophic security flaws and pose catastrophic risks, and that duality is precisely why the company argues for stronger, government-backed oversight.

That argument is spelled out in Anthropic's Advanced AI Framework, a policy document the company has published alongside its funding activity. The framework calls for government verification of the safety claims AI companies make about their own products, along with civil penalties for firms that fail to follow safe practices. It also proposes giving regulators authority to delay or outright block the deployment of AI models judged to carry catastrophic risk, a notably interventionist stance for a company that itself builds and sells frontier models.

Anthropic's willingness to fund political advocacy at this scale sets it apart from peers that have generally kept a lower profile on state-level AI legislation. Combining that spending with evidence of Claude Mythos's vulnerability discoveries appears designed to make a concrete case: the same systems capable of finding critical software flaws across every major OS and browser are advanced enough to warrant binding safety verification, not just voluntary industry commitments. Whether other AI labs follow with comparable financial commitments to policy advocacy, or push back against the regulatory model Anthropic is proposing, will likely shape how state and federal transparency rules evolve from here.