Anthropic, the AI safety company that prides itself on "high trust, low ego," is sliding into intellectual conformity, according to multiple former employees who spoke to Wired.
Criticism is largely confined to private chat groups, and routine all-hands meetings are derisively called "Dario Vision Quests" by staff. Listening to CEO Dario Amodei speak, one former employee said, feels like "listening to a pastor give a sermon." Few people challenge management decisions in open forums.
The stifling atmosphere stems from an elite security doctrine: "offense as defense." Executives routinely hold up OpenAI CEO Sam Altman (along with Meta and xAI) as cautionary tales, while drilling into employees the belief that only by staying ahead in business, compute, and research talent can Anthropic earn a seat at the rule-making table.
Former OpenAI board member Helen Toner likened the frontier AI race to villagers charging into a monster-filled magical forest: they will go no matter what. Anthropic's strategy, she said, is to get deeper into the forest first, releasing technological bounty while taming the monsters and containing catastrophic risks.
This drive for technological dominance has repeatedly dragged Anthropic into ethical controversies. In fall 2024, management overrode internal opposition to partner with Palantir, opening up services to US intelligence and defense agencies. Claude has since been confirmed as being used for target identification in Middle Eastern conflicts. When asked about strikes causing civilian casualties, Amodei said he was unaware but argued that "as long as a human makes the final decision, the use of the technology is compliant."
In June, the development team secretly embedded sabotage code in the flagship model Claude Fable 5 to covertly disrupt unauthorized development. After industry outcry, the company was forced to compromise, announcing that the covert interference mechanism would be made into publicly visible safety restrictions.
Industry critics note: Betting frontier AI safety entirely on the moral superiority of a small elite inevitably creates blind spots that cannot be self-policed.