Anthropic CEO Dario Amodei published an essay on June 10 calling for the federal government to have explicit authority to block, or even reverse, the deployment of AI models deemed too dangerous for public safety.

What Anthropic is actually asking for

The company’s recommendations center on three pillars. First, mandatory independent safety testing for advanced AI systems, specifically those classified as frontier models. Second, a comprehensive federal framework designed to manage what Anthropic calls “catastrophic AI risks.” Third, and perhaps most politically charged, a warning against preempting state-level AI regulations unless Congress passes robust federal legislation first.

That last point is a direct shot across the bow of the Trump administration, which has been pushing for a federal legislative framework that would effectively override state regulations.

The June 10 essay didn’t arrive in a vacuum. Just five days earlier, on June 5, Anthropic released a separate statement urging AI labs to consider verifiable pauses in development. The argument is straightforward: if self-improving AI systems start escalating beyond what society can safely manage, the industry should have mechanisms to pump the brakes.

The backstory and the tensions

Anthropic was founded in 2021 by former OpenAI leaders, and safety has been its brand identity from day one. Dario Amodei and his sister Daniela left OpenAI specifically because they wanted to build an AI company with alignment and safety baked into its DNA, not bolted on as an afterthought.

Tensions between Anthropic and the Department of War over AI model safeguards have been an ongoing issue since early 2026. The details of those negotiations remain closely held, but the dynamic is clear: Anthropic wants guardrails, and certain corners of the government want access to frontier AI capabilities without too many strings attached.

It’s worth noting that none of Anthropic’s announcements or related discussions referenced digital assets, cryptocurrency tokens, or blockchain technologies. This is a pure AI governance play, no Web3 angles to be found.

What this means for investors and the AI landscape

If Congress actually follows through on anything resembling Anthropic’s recommendations, mandatory independent safety testing for frontier models would mean enormous compliance costs. Think of it like pharmaceutical drug trials, but for software that can write code, generate synthetic media, and potentially assist in bioweapons research.

That kind of regulatory burden tends to favor incumbents. Companies like Anthropic, OpenAI, and Google DeepMind have the resources to navigate compliance regimes. Smaller startups building on the frontier would face a very different calculus, potentially creating barriers to entry that consolidate market power among a handful of established players.

The irony of Anthropic’s position is worth sitting with. A company racing to build increasingly powerful AI systems is simultaneously asking the government for the authority to stop companies, including potentially itself, from releasing those systems.