On July 21, Anthropic announced it’s donating another $20 million to Public First Action, bringing the total to $40 million. Public First Action is a non-partisan organization that educates the public about AI and works with Republicans, Democrats and Independents on sensible guardrails. Both donations are explicitly for education and policy work only — not to campaign for any candidate.
Why now?
Anthropic gives a remarkably direct answer. The models are getting better at a blistering pace — and that isn’t only good news. Anthropic points to Claude Mythos Preview, which earlier this year discovered thousands of high-severity vulnerabilities, including some in every major operating system and browser. It released the model only to a limited set of cyber defenders through Project Glasswing, so those weaknesses could be fixed before anyone exploited them. In the wrong hands, the argument goes, models like this could threaten critical systems — from hospitals to the power grid.
The core message: governments need time to build the capacity to capture the benefits while containing the risks. And that’s exactly why they need to start now.
What Anthropic is asking for
Transparency alone is no longer enough, Anthropic writes — even though it has supported new transparency laws in several US states. What’s needed are measures with real teeth: governments should be able to verify companies’ safety claims, enforce safe practices through civil penalties, and, in the worst case, slow or block the deployment of models that pose a serious risk of catastrophic harm.
My take
You shouldn’t read this naively. An AI lab pouring money into political education is also pursuing its own interests — and Anthropic is one of the companies that could well benefit from stricter rules for everyone. That belongs in the picture too, honestly.
Still, I find the underlying stance consistent. Anthropic has said the same thing for years — the models are getting dangerously powerful, fast, and policy is lagging behind. A company voluntarily campaigning for tighter oversight of itself is at least unusual. Whether $40 million to an education group makes the difference remains to be seen. But the question behind it is real: who actually sets the rules before the technology sets the facts?
Sources: