Sam Altman warns AI control could concentrate in too few hands
OpenAI's CEO used a weekend podcast to stake out the industry's defining question — not whether AI becomes powerful, but who gets to decide what it does to everyone else.
Sam Altman's biggest fear for artificial intelligence isn't a runaway superintelligence — it's a future where a small group of companies or people decides how it reshapes your life. Speaking with podcaster David Senra on Sunday, the OpenAI chief warned that ordinary people risk losing their voice entirely as the technology advances, and argued AI's direction should be steered by broad social choices rather than whatever path individual companies prefer. "The right approach is for people to deeply control the future," he said. "Society and the model will evolve together."
His sharpest point was saved for the safety-first camp. Altman said AI escaping human control genuinely worries him — but he argued the fear itself could cause the concentration problem, because people "so nervous about the magnitude of those risks" will push to trade "a lot of liberty for safety," handing authority to whoever positions themselves as the protector. It reads as a direct swipe at rival Anthropic: CEO Dario Amodei has spent months urging governments to be able to block dangerous model capabilities, and Anthropic declined to endorse open-weight releases even as Chinese labs use open models to match Western frontier performance at a fraction of the running cost. We covered Amodei's side of that argument in July — Amodei: open weights are 'nowhere near a sufficient solution'. There's an awkward wrinkle for Altman, though: OpenAI's own frontier models remain fully closed, with last year's gpt-oss releases the exception rather than the rule.
The interview's more credible moment was a timing confession. Altman admitted he expected GPT-4-era models to disrupt the economy almost immediately after their 2023 debut — software businesses above all — and that he was wrong, because organizations can't absorb new capabilities faster than they can rework their software, data access, compliance processes and budgets. He framed the lag as a blessing in disguise: a slower, smoother transition means less economic and social whiplash along the way.
Our take: the liberty-versus-safety framing doubles as competitive strategy — OpenAI is fighting regulation on one flank and Chinese open-weight labs on the other, and "let society decide" conveniently serves both fights. But the humility about timelines is worth more than the ideology. An industry that habitually overpromises speed needs more executives willing to say plainly that they misjudged.
What to watch: whether OpenAI's next frontier release ships with downloadable weights. If concentration really is the enemy, open weights would be the loudest way to prove it.
If AI power does concentrate, who should hold the keys — markets, governments, or the labs themselves? Tell us in the comments.
