It’s Time to Investigate the AI Labs

Over the past few months, OpenAI and Anthropic have managed to look less like research labs and more like political actors with demo budgets. First came OpenAI’s carefully staged warnings about how powerful and troubling its agent systems have become. Then Anthropic employees started openly debating whether these systems could actually end human life. That is not normal corporate behavior, even by Silicon Valley standards.

The latest move was Anthropic CEO Dario Amodei’s letter, “We Must Pace the Frontier.” It lays out the harms his company’s research could cause, then lands on a familiar conclusion: the technology should be built “in the right way,” with government slowing potential competitors while the frontier labs keep leading the charge. Sam Altman quickly backed the idea. The message was hard to miss: let us steer, or else.

That pitch has not exactly landed as intended. Instead of persuading everyone to trust the labs’ moral clarity, it has triggered a more basic reaction: what exactly is going on in these places? The source of the discomfort is not abstract AI hype. It is a narrow set of risky experiments, run mainly by the frontier labs, and the problems they keep producing.

Last Thursday, the writer behind the piece took that question to the New York Times and called on Congress to launch a public fact-finding effort. The point was not to debate “AI” as if it were one smooth, inevitable thing. It was to ask what specific systems OpenAI and Anthropic are building, how they are doing it, and what they think they are for.

The other targets are internal safeguards and the ideology driving the work. OpenAI has already disclosed a long chain of unauthorized hacking attacks by its autonomous agents, which raises an awkward follow-up: why weren’t these efforts stopped sooner, and why wasn’t criminal liability even considered if the systems were expected to commit crimes? And behind all this sits a broader worry that both labs are becoming more reckless because they see collateral damage as the price of saving humanity. Congress, the argument goes, should stop letting a small group of private companies set the terms of the whole conversation.