Anthropic, the maker of Claude, has become isolated in several major AI policy disputes in the US and Europe while remaining commercially strong, TNW reported on July 29. The company, described in the article as the most valuable startup in the world and the most isolated frontier AI lab, has taken stricter positions on open weights, model safety, defense use and distillation than many rivals.

The sharpest split came last week, when Anthropic was the only frontier lab that declined to sign an open letter urging Washington not to restrict open-weight models (models whose parameters are publicly released). Nvidia CEO Jensen Huang led the letter, and Google and OpenAI signed it. Anthropic CEO Dario Amodei said the company had not called for banning open models and called weaker ones “a public good,” but he maintained that advanced chips reaching authoritarian states need tighter control.

The article said the pattern extends to other fights. The Pentagon blacklisted Anthropic in February after disagreement over whether Claude could be used for surveillance or autonomous weapons, and one senior defence official said Friday there was “no AI company more hostile to the warfighter.” Anthropic has also backed state AI laws and proposed an FAA-style regulator to vet models before release. When the White House flagged a security flaw this summer, Anthropic disputed its severity; the dispute helped trigger export controls that forced two of its models offline for nearly three weeks.

Anthropic has accused Chinese labs of using Claude outputs for “industrial-scale” distillation (training models from another model’s outputs), according to the New York Times. Critics called distillation normal in AI development and pointed to Anthropic’s $1.5bn copyright lawsuit settlement over pirated books used to train its own models. Anthropic still joined more than 1,100 rival-lab employees this week in urging Washington to “deliberately pace” frontier AI, and the article said its models top most independent benchmarks, enterprises pay premium prices, and its valuation reached $965bn in May ahead of a possible IPO.

Anthropic, the maker of Claude, has become isolated in several major AI policy disputes in the US and Europe while remaining commercially strong, TNW reported on July 29. The company, described in the article as the most valuable startup in the world and the most isolated frontier AI lab, has taken stricter positions on open weights, model safety, defense use and distillation than many rivals.

The sharpest split came last week, when Anthropic was the only frontier lab that declined to sign an open letter urging Washington not to restrict open-weight models (models whose parameters are publicly released). Nvidia CEO Jensen Huang led the letter, and Google and OpenAI signed it. Anthropic CEO Dario Amodei said the company had not called for banning open models and called weaker ones “a public good,” but he maintained that advanced chips reaching authoritarian states need tighter control.

The article said the pattern extends to other fights. The Pentagon blacklisted Anthropic in February after disagreement over whether Claude could be used for surveillance or autonomous weapons, and one senior defence official said Friday there was “no AI company more hostile to the warfighter.” Anthropic has also backed state AI laws and proposed an FAA-style regulator to vet models before release. When the White House flagged a security flaw this summer, Anthropic disputed its severity; the dispute helped trigger export controls that forced two of its models offline for nearly three weeks.

Anthropic has accused Chinese labs of using Claude outputs for “industrial-scale” distillation (training models from another model’s outputs), according to the New York Times. Critics called distillation normal in AI development and pointed to Anthropic’s $1.5bn copyright lawsuit settlement over pirated books used to train its own models. Anthropic still joined more than 1,100 rival-lab employees this week in urging Washington to “deliberately pace” frontier AI, and the article said its models top most independent benchmarks, enterprises pay premium prices, and its valuation reached $965bn in May ahead of a possible IPO.