Dario Amodei, chief executive of AI safety company Anthropic, is calling for sweeping new government powers to halt dangerous artificial intelligence systems before they reach the public, arguing that current policy frameworks are dangerously out of step with the speed of AI development.
In an essay released Wednesday, Amodei laid out a detailed regulatory agenda that goes far beyond anything currently debated in Washington. He argues that frontier AI models should face mandatory testing for risks tied to cybersecurity, biological weapons, loss of control, and automated research capabilities. If a system fails these tests, he contends, the government should have the authority to block its release or reverse its deployment as a public safety threat.
The framework mirrors how governments regulate other high-risk technologies. "Like cars, airplanes, or drugs, AI regulation should require frontier models to go through rigorous testing and auditing," Amodei writes. He criticizes existing transparency rules as insufficient and calls for "more serious and binding regulation."
Amodei also tackled economic fallout from AI, proposing a broader policy agenda that includes wage insurance, tax incentives for job retention, capital accounts, and possibly universal basic income. He argues that job displacement from AI could exceed disruption from previous technological shifts and may be longer-lasting, though he frames public resistance to data center construction as a symptom of deeper economic anxiety rather than a fundamental objection to the technology itself.
The CEO's pitch arrives as the Trump administration has issued an executive order on AI. Amodei suggests that order should go further with the mandatory testing requirements he outlined. He also calls for regulatory modernization at agencies like the FDA to accelerate approval of AI-discovered drugs, while backing domestic bans on fully autonomous weapons and continued coordination among democratic nations on AI governance.
The proposal is likely to draw sharp criticism on multiple fronts. Tech industry rivals may argue that Anthropic is designing a regulatory moat to entrench its competitive position. Critics already skeptical of AI doomsaying will see it as corporate fearmongering dressed up as policy. Some libertarian voices will reject the notion of AI blocking powers as government overreach.
Amodei preempted at least part of this pushback by emphasizing that many of his proposals, from job displacement programs to export controls on chips, carry common-sense appeal across the political spectrum. He also stressed his optimism about finding workable solutions, noting that momentum exists for addressing multiple policy challenges simultaneously.
The timing reflects a genuine inflection point. Anthropic recently released its Mythos power model, and Amodei warns that biological risks and "serious AI autonomy risks" could materialize sooner than many expect. "Policy and especially legislation moves very slowly," he writes, while "AI is advancing at a lightning pace." The gap between those two trajectories, he argues, demands urgent action before the window closes.
Author James Rodriguez: "Amodei's playing a calculated game here, combining legitimate safety arguments with economic protections that benefit his industry, but his core point that regulators are asleep at the wheel deserves real scrutiny from policymakers."
Comments