The AI industry's leading firms are asking the public to trust them at precisely the moment their own actions are giving people reasons not to. An Anthropic researcher resigned this week warning that the company is "racing straight to self-improving superintelligence and gambling with our lives," and the company's own alignment lead co-signed the message rather than walking it back, as TechCrunch reported. At the same time, the same company reportedly faces allegations of surveilling AI activists, OpenAI has stopped taking Pro subscriptions because it cannot meet demand for its Astra product, and investors are racing to pump money into robot-training data. The thread is a widening gap between what AI leaders say about safety and what they actually do.
The Warning From Inside
The most striking detail in the Anthropic story is not the warning itself. The industry has flirted with doomer rhetoric before, and a resigning researcher's X post would ordinarily be filed away as an internal dispute. What makes this different is the co-signature. The alignment lead choosing to endorse the message rather than distance the company from it signals that the concern is not confined to one departing employee. It suggests an institutional split, not a personal grievance. The timing compounds the signal. With Anthropic reportedly preparing for an IPO, the company has an obvious incentive to project stability and confidence to public-market investors. Instead, its own staff are publicly describing the core mission as a gamble. That is a governance problem before it is a public-relations one. For US technology companies preparing to sell shares to ordinary investors, a co-signed internal warning about existential risk is exactly the kind of material disclosure that invites scrutiny from regulators and prospective shareholders alike.
The Precrime Allegation
Running alongside the safety rhetoric is a separate and arguably more damaging story. A recent report alleges Anthropic is finding ways to monitor activists and surveil dissent, as CNET reported. The framing invites an obvious comparison to pre-crime: treating people as future threats based on their political activity rather than any wrongdoing. Whatever the merits of the underlying reporting, the allegation lands in the same news cycle as the company's own employees warning about its direction. That combination matters for American consumers and workers. AI firms increasingly position themselves as neutral infrastructure, but the same systems capable of monitoring dissent are the ones being embedded in workplaces, schools, government services and consumer products. If the companies building these systems are themselves suspected of watching their critics, the public has little reason to extend them the benefit of the doubt on data handling, bias, or safety more broadly.
Demand Outruns Everything
OpenAI's decision to pause Pro subscriptions because of Astra demand, as TechCrunch reported, reads on its face as a good-news story. It is also a stress signal. The company said Pro subscriptions put the most strain on its systems, so it is halting sign-ups while adding capacity. The immediate consumer consequence is straightforward: US customers who want to pay for the top tier cannot, at least for now. The deeper consequence is about reliability. A company that cannot accept paying customers is a company whose infrastructure has not kept pace with its commercial ambitions. That undercuts the reassurance that AI systems are dependable enough to embed across the economy. When capacity constraints force a major provider to close the door on revenue, the question of whether the industry can responsibly manage the risks its leaders themselves describe becomes harder to answer in the affirmative.
