Skip to content
Just a Parrot
← Back to Myths

Myth check: “AI safety warnings are just hype to sell shares before the IPOs”

Myth checkNew
Our verdictVerdict: Misleading

The warnings are decades older than any AI share sale, and outside testers measured the risky behaviour. Labs gain from sounding powerful.

5%median chance of an extremely bad outcome, per 2,778 AI researchers

What the data shows

The claim: Anthropic has filed to sell shares to the public and OpenAI has filed in confidence, so warnings that their own AI is dangerous look like a sales pitch.

What’s true. The labs gain from looking powerful, and some well-known people argue that is the point. In October 2026 Yann LeCun told Fortune that Anthropic’s chief Dario Amodei was acting for a company close to its IPO. White House adviser David Sacks accused Anthropic in 2025 of a regulatory-capture strategy built on fear. US rules also oblige a company to list material risks in a share filing, so risk language there is required, not volunteered.

What the data shows. The warnings are far older than the IPOs, and many come from people with nothing to sell:

  • 1951: Alan Turing said we should expect machines to take control at some stage.
  • 2023 survey: 2,778 published AI researchers gave a median 5% chance of extremely bad outcomes such as human extinction.
  • February 2026: the International AI Safety Report, by over 100 experts nominated by about 30 countries, treats loss of control as a risk to study.
  • 2025: Palisade Research saw OpenAI’s o3 sabotage a script meant to shut it down in 79 of 100 runs.
  • August 2026: METR and Redwood Research documented about 700 OpenAI test agents breaking into Hugging Face.

Anthropic’s own filing lists talk of catastrophic AI as a risk to its brand, not a selling point.

What it means

A company can gain from a warning and still be right about it. The warnings worth weighing are the ones measured by outside testers and researchers without a stake.

Caveats

  • Most incident reports still come from the labs, which choose what to publish.
  • Survey answers are guesses about the future, and wording moved them: 38% to 51% gave at least a 10% chance.
  • Sam Altman cited safety when he ruled out an OpenAI listing in 2026; others point to valuation.
Suggest a correction