Anthropic Built Its Identity on Being the Anti-OpenAI. That Positioning Is Now Under Pressure.

The story of Anthropic's founding has always been framed as ideological: a group of researchers who left OpenAI because they believed the lab was moving too fast, with too little care for the consequences. That framing is accurate as far as it goes. But a detailed account of the split, drawn from Keach Hagey's reporting on the people who built the current AI industry, reveals that the ideological story coexists with something more human: personal dynamics, power struggles, and the accumulating grievances that precede most institutional breakups.

The internal frame Anthropic employees use to describe OpenAI is pointed: the tobacco industry. The analogy is precise. Tobacco companies knew about the health consequences of their product for years before that knowledge became public liability. The concern within Anthropic, as articulated by employees and captured in Hagey's reporting, is that OpenAI is following the same pattern — building capability, understanding risk, and proceeding anyway.

The Pentagon Moment

The frame crystallized around a specific event. When OpenAI accepted a Pentagon contract for surveillance and autonomous weapons applications that Anthropic had publicly declined, Anthropic CEO Dario Amodei responded internally with an assessment of Sam Altman that went beyond policy disagreement. Amodei called Altman "mendacious" and described the Pentagon decision as reflecting "a pattern of behavior that I've seen often from Sam Altman." The language is unusually personal for a professional context — and unusually candid for a CEO whose company competes directly with the target of the assessment.

Amodei has publicly warned that AI companies that fail to disclose safety risks could end up "in the same position as cigarette manufacturers and opioid producers who knew about dangers but stayed silent." The framing positions Anthropic's transparency commitments not just as policy preferences but as moral obligations with legal and reputational analogues.

The Safety Pledge Revision

Against that backdrop, one recent Anthropic decision deserves more attention than it has received. The company replaced its binding internal safety pledge — a commitment that constrained certain categories of development — with a non-binding framework that the company acknowledges can be revised as competitive conditions change. The revision was made quietly. The consumer wave driving Claude's growth was not.

The tension is structural: the identity that is driving Anthropic's consumer momentum is built on the claim that it takes safety more seriously than its competitors. If commercial scale erodes the commitments that underwrite that claim, the consumer loyalty built on it becomes fragile in precisely the way that scale-driven AI safety concerns were always predicted to be.