TL;DR:

• Anthropic chief executive Dario Amodei told the Financial Times he wants AI regulated “the way you regulate cars and aeroplanes” — mandatory, third-party-assessed, and built around safety standards.
• He defended Anthropic’s refusal to allow Pentagon domestic-surveillance uses despite a presidential rebuke, and said open-source and Chinese labs are six to twelve months behind Claude Mythos.
• Amodei proposed the Frontier Model Forum as a vehicle for industry-wide safety benchmarks, and said biosecurity risks of comparable scale could emerge within a year.

In a wide-ranging Lunch with the FT interview in San Francisco, Amodei argued that the Mythos release — and the regulatory scramble it triggered — has changed the policy conversation. He pointed to the Frontier Model Forum, an industry body, as a possible setter of the “brakes, airbags and seatbelts” standards he wants mandatory for frontier models.

The doctrine after Mythos

Amodei’s position has been evolving under pressure from both directions. The White House labelled Anthropic’s leadership “radical leftwing nut jobs” after the company refused to permit Pentagon use of its models for domestic mass surveillance or fully autonomous weapons, and the Department of Defense classified Anthropic a “supply chain risk” — a decision the company is contesting in court. On the other flank, tech commentator Ben Thompson has accused Amodei of “disaster-porn-as-marketing”.

The FT interview places Amodei’s regulatory push in a third category: he is now pre-committing to external oversight while Anthropic retains a technology lead. He predicts open-source and Chinese labs will replicate Mythos’s vulnerability-discovery capabilities inside twelve months, and is directing $20 million of political donations through the Public First Action Super PAC to support state-level AI safety legislation in the US.

UK relevance

Amodei’s cars-and-aeroplanes framing is more interventionist than the UK government’s preferred “pro-innovation” regulatory stance and the approach Chancellor Rachel Reeves outlined around the £500 million Sovereign AI Fund. It sits closer to the EU AI Act’s structural approach than to Westminster’s light-touch defaults, and is likely to sharpen the regulatory divergence already visible between London, Brussels and Washington. UK banks, which Anthropic will open Mythos access to within the week per The Guardian, are effectively the first large non-US commercial users exposed to the model under whichever framework emerges.

Looking forward

The practical question now is whether the Frontier Model Forum can move from industry self-governance toward something with teeth before Amodei’s predicted biosecurity capability jump arrives. His prediction that such risks could materialise within six to twelve months sets a clock on policy-making that UK regulators — still working through Mythos cyber-risk assessments — will struggle to match at current cadence.