Anthropic CEO Dario Amodei published an essay on June 10 titled “Policy on the AI Exponential,” arguing that AI is advancing far faster than governments can legislate. He opens with a Lord of the Rings image, comparing slow political institutions to Treebeard, the ponderous talking tree. His core claim is that if scaling laws hold for another year or two, the world could reach what he calls “a country of geniuses in a datacenter.” Alongside the essay, Anthropic released a legislative proposal on frontier model testing and a framework for job displacement, pledging substantial financial backing. Some coverage put that funding at about $350 million.
The headline shift is from transparency to binding rules. For three years Anthropic argued that companies should disclose safety procedures and test results while legislation waited. It backed transparency laws in 2025, including California’s SB 53, New York’s RAISE Act and Illinois SB 315. Now, Amodei says, the risks are concrete. He points to Claude Mythos Preview, which he says showed frontier models can pose serious cybersecurity threats to finance, infrastructure and national security.
His proposal models AI oversight on the Federal Aviation Administration. Key elements:
- Mandatory third-party testing for models above a compute threshold, covering cybersecurity, biological weapons, loss of control, and automated R&D that could accelerate those risks
- Government power to block or reverse a model’s release if it fails, scoped to those four risks with safeguards against political favoritism
- Evaluation by a government agency or by private, government-authorized assessors, an approach he calls regulatory markets
- Strong security for model weights, regular red teaming, and prompt reporting of safety incidents
The essay spans five areas. On the economy, Amodei warns AI could lock in rapid growth alongside extreme inequality, and calls enduring job loss undesirable but possible. He floats wage insurance, retention tax incentives, training grants and, if displacement proves large, long-term income support such as universal basic income funded by taxes on AI firms or higher capital gains rates. On civil liberties, he warns powerful AI could become a tool of autocracy and surveillance, urging limits on autonomous weapons and the closing of a data-broker loophole. On geopolitics, he likens AI to nuclear weapons and calls for a coalition of democracies that shares chips and standards while denying them to rivals.
The Bigger Picture
- The essay signals the end of a light-touch posture from a leading AI developer. It lands as Anthropic and rivals move toward potential IPOs.
- For enterprises, an FAA-style block or recall could cut access to a flagship model, strengthening the case for multi-vendor strategies.
- Amodei rejects the idea that AI fear is a marketing problem, framing public concern as democratic accountability working as intended.
What Came Before
Anthropic’s earlier strategy stressed preserving optionality: transparency rules, chip export controls and data on AI’s labor effects. Amodei says those steps were all that seemed possible when AI’s risks were still hypothetical. He credits the Trump administration’s executive order for moving incrementally toward a larger federal role, while arguing his plan goes further. He also notes that voluntary frameworks like Anthropic’s Responsible Scaling Policy taught the company that rigid, pre-written checklists can miss the biggest risks. The timing is pointed: the essay arrived very close to the launch of Claude Fable 5, a juxtaposition some commentators were quick to highlight.
