Ban My AI! Anthropic Boss Demands Aviation-Style State Control

The sudden change of heart materialized in a massive manifesto by Anthropic chief Dario Amodei, published right after his company dropped its latest Claude Fable 5 model. He is now pushing for a regulatory regime where state agencies can literally veto or recall high-end AI models before they ever see the light of day, much like the FAA grounds unsafe airplanes.

Under this plan, any AI model crossing a certain computing threshold must undergo audit by a government-approved third party across four apocalyptic risk categories: cyber warfare, biological weapons creation, rogue autonomous behavior, and AI self-replication. It turns out the old corporate song of "just make us write transparency reports" is no longer enough now that these models are actually getting powerful.

To prove his point, Amodei pointed to their own Claude Mythos Preview, claiming it already disrupted global cybersecurity. The proposed rules would force companies to secure their model weights like military secrets, which, by pure coincidence, is incredibly easy for multi-billion-dollar giants and virtually impossible for the open-source community.

Knowing that robots replacing humans makes for bad PR, Anthropic is throwing a $350 million bone to social initiatives, suggesting wage insurance for displaced workers and even an AI-taxed Universal Basic Income. They also want to ban fully autonomous police weapons and guarantee that any citizen facing state prosecution gets access to AI that is just as powerful as the government’s prosecutorial algorithm.

It is a masterclass in regulatory capture, disguised as digital humanitarianism. By lobbying for rules that effectively outlaw open-source AI weights—since a published open model can never be "recalled"—the industry leaders are trying to pull up the ladder behind them under the guise of saving humanity from the very monsters they are building.

Source: Dario Amodei's Blog