As AI Labs Warn of Danger, Critics Ask What They Stand to Gain

Why would the companies building the most advanced AI systems warn that those systems are dangerous? An Associated Press analysis published by ABC News on September 27, 2026, took up that question, and found experts who think the answer may involve politics and money as well as safety.

The AP noted that the CEOs of Anthropic and OpenAI have publicly called their own models dangerous and urged regulation and independent testing before release. In a rare show of unity, they have laid out alarming scenarios in essays, social media posts and speeches at the United Nations.

Setting the terms

Experts, analysts and former government evaluators told the AP that, with this rhetoric, the companies appear to be seeking public favor and setting the terms of their own oversight. The report tied the timing to two events: the US midterm elections and planned stock listings, for which the labs and their investors need fresh capital.

One analyst, PitchBook senior research analyst Harrison Rolfes, told the AP that calls for caution seem meant to curry favor with investors ahead of public offerings and the midterms, and argued that positioning as the safest bet can help the largest labs build “a wall or a moat” against smaller competitors.

The agency they are not asking for

A central point of the analysis concerns the federal Center for AI Standards and Innovation, or CAISI, which already evaluates some advanced models. The AP reported that the companies are not pushing for more oversight from that agency. Conrad Stosz, who previously led CAISI, told the AP the labs are instead vowing to create their own auditing parameters and choose particular evaluators to grade them.

Stosz, now head of governance at the research lab Transluce, said many evaluators want deeper access to the labs, but that it is ambiguous what “embedded evaluators” would mean and whether they could investigate without losing their independence.

The White House position

The report also noted that President Trump has rejected new AI regulation, dismissing talk of risks to humanity as a “HOAX.” That leaves the companies’ own commitments as the main form of oversight on the table, which is exactly what critics quoted by the AP worry about.

The companies pushed back. An Anthropic spokesperson told the AP the company has been calling for regulation for several years. OpenAI spokesperson Liz Bourgeois pointed to the company’s recent pause in training its most advanced models (read our report on the pause) and said, “People want to know AI is being developed safely, and that starts with what companies like ours do ourselves.”

What came next

Two days later, the White House and leading AI companies signed a voluntary accord built around internal controls and an external auditor chosen by each company, putting the AP’s questions about self-designed oversight directly to the test; see our report on the accord.