Anthropic researcher resigns and warns company is racing toward self-improving superintelligence

An Anthropic researcher announced his resignation this week and, in a post on X, charged that the company is “racing straight to self-improving superintelligence and betting our lives on it.” The head of the company’s alignment team signed the resignation letter rather than deny or soften the claim. The warning arrives as Anthropic is, according to reports, preparing for an initial public offering — a fact that changes the weight of the allegations compared with earlier industry “doom” warnings.

In a new episode of "Equity", the TechCrunch podcast, hosts Kirsten Korosec, Anthony Ha and Sean O’Kane dissect the resignation and what it reveals about the race toward models of ever-greater capability. The episode also covers Apple’s first major event under new chief executive John Ternus, a one billion dollar raise for Stokes Space, a two billion dollar round for Cognition at a $48 billion valuation, and a $50 million seed for floating nuclear reactors to power AI data centers.

Apple’s event centered on AI embedded in hardware — “always-on” features in the Watch and AirPods — more than on the new foldable iPhone. That approach signals a shift from a flashy product to infrastructure that runs in the background all the time, and it is exactly the kind of integration that creates deeper dependence on the models and the infrastructure behind them.

The Stokes Space raise reflects a launch market that has suddenly opened up as SpaceX cuts back Falcon flights. At the same time, the massive investment in Cognition, a code-generation startup that has yet to prove a product at broad scale, shows how much venture capital is willing to bet on the assumption that there is still enormous room for automated code writing. The seed round for floating reactors illustrates the pressure to deliver reliable, always-available power for the data centers running these models.

The alignment lead’s signature on the resignation letter is not merely symbolic. It signals that inside the company there are people who believe the current pace, and the pressure to show growth ahead of an IPO, is overtaking the checks needed before releasing systems capable of improving themselves. Without public benchmarks or peer review of the claims about “self-improving superintelligence,” all that remains is the word of former employees against the company’s financial interest.