Anthropic CEO slams OpenAI’s Pentagon deal as ‘safety theatre’

Anthropic CEO Dario Amodei

• Anthropic criticizes OpenAI's defense deal over ethical concerns
• OpenAI allows "all lawful use," Anthropic wanted stricter limits
• ChatGPT uninstalls surged after OpenAI's Pentagon agreement

Did our AI summary help?

The rivalry between Anthropic and OpenAI has moved beyond product launches and model benchmarks. It is now about who gets to define the moral perimeter of artificial intelligence — and who is willing to redraw it.

In a staff memo reported by The Information, Anthropic co-founder and chief executive Dario Amodei sharply criticised OpenAI’s new agreement with the US Department of Defense. He described the company’s public framing of the deal as “safety theatre” and accused its leadership of misrepresenting the substance of negotiations.

The dispute centres on a familiar but combustible phrase: “any lawful use.”

Anthropic, which already holds a $200 million defence contract, had been in talks with the Pentagon over expanded access to its systems. According to the memo, those talks broke down after the military declined to explicitly rule out domestic mass surveillance and autonomous weapons applications. Anthropic wanted written assurances. The Department would not provide them.

OpenAI, by contrast, proceeded with its own agreement. In a blog post, the company said its contract permits the use of its systems for “all lawful purposes” while stressing that mass domestic surveillance is illegal and not covered. It added that it had ensured those limits were made explicit in the contract language.

For Anthropic, that was not enough. The concern is not what is illegal today, but what might be redefined tomorrow. Laws change. Administrations change. The phrase “lawful use” is elastic, particularly in matters of national security.

That tension is amplified by the political climate. The Department of Defense, operating under the Trump administration — sometimes referred to internally as the Department of War — has taken a more expansive view of executive authority in other domains. Critics argue that relying on present-day legal definitions offers limited long-term reassurance.

Amodei’s memo suggests the disagreement is philosophical as much as contractual. He claimed OpenAI prioritised internal optics and employee management, while Anthropic focused on setting hard constraints around potential abuse. He also dismissed OpenAI chief Sam Altman’s public positioning as conciliatory, calling it misleading.

The clash is not confined to executive correspondence. Public reaction appears to have tilted in Anthropic’s favour. ChatGPT uninstalls reportedly surged nearly 300 per cent following news of the defence deal, and Anthropic’s app ranking climbed sharply in the US App Store. Whether that spike represents lasting sentiment or a temporary protest remains to be seen.