Anthropic pretraining researcher quits, warns of reckless AI superintelligence race

A researcher posting as @hilbertspaess announced their resignation from Anthropic on September 9, 2026, capping a three-year stint doing pretraining research split between OpenAI and Anthropic. In a short thread, they accused both labs of behaving irresponsibly, charging that the industry’s leading players are sprinting toward self-improving superintelligence and, in their words, gambling with people’s lives.

The post offers no technical evidence or specifics — it is a personal statement of conscience rather than a documented disclosure — but it lands with weight given the author’s claimed vantage point inside the pretraining teams at the two most prominent frontier labs. It slots into a recurring pattern of high-profile safety-motivated departures from these organizations, where insiders publicly break with their employers over the pace and perceived recklessness of capability development.

The claims are unverified and circulating on social media, so readers should treat them accordingly; the account’s identity and the details of any internal disagreement aren’t independently confirmed. Still, the resignation adds another data point to the ongoing tension between competitive pressure to ship ever-more-capable models and the safety community’s warnings that the field is moving faster than its ability to control what it builds.