Anthropic has done something genuinely clever: it has made being cautious into a product differentiator. Every time Claude declines a request that GPT-4o handles without complaint, Anthropic can frame that friction as evidence of responsibility rather than limitation. That framing has held together remarkably well - but it’s getting harder to accept at face value.
The company was founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei, explicitly around safety concerns. That origin story is real, and the research Anthropic publishes on interpretability and alignment is substantive work. Constitutional AI, the method they developed for training Claude to evaluate its own outputs against a set of principles, is a genuine contribution to the field. None of that is in dispute.
What’s worth scrutinising is how the safety posture functions commercially. Anthropic operates in a market where OpenAI has name recognition, Google has infrastructure, and Meta is giving models away for free. Competing on raw capability against those players is expensive and uncertain. Competing on trustworthiness - especially in enterprise sales, where legal and compliance teams have actual veto power - is a more defensible position. The pitch writes itself: we’re the one AI company that will tell you when our model shouldn’t be used.
That pitch is working. Anthropic has secured major contracts with Amazon and raised funding at valuations that would have seemed implausible for a company its size a few years ago.
The problem with safety as brand
When safety is also your go-to-market strategy, the two start to contaminate each other in ways that are difficult to untangle from the outside. Every policy decision - what Claude will and won’t do, which capabilities get restricted, how the model handles ambiguous requests - carries both a technical justification and a reputational implication. You can’t easily tell which one is driving the choice.
This isn’t unique to Anthropic. OpenAI faces the same tension. But OpenAI has mostly stopped pretending otherwise, pivoting visibly toward capability and commercial scale. Anthropic still presents itself primarily through the lens of safety research, which invites more scrutiny when the business interests align a little too neatly with the ethical stances.
None of this means the safety work is cynical or fake. Researchers at Anthropic are doing things that matter. But the company has made safety its narrative center of gravity, and narratives that convenient deserve pressure-testing - not because they’re wrong, but because the incentive to keep repeating them doesn’t require them to stay true.