Altman: AI could go 'really wrong' in two ways, and we can't get it back

The two ways Altman says AI could go wrong

Speaking on the Figuring Out With Raj Shamani podcast, Altman sketched out two opposite failure modes.

"One is that there's a loss of control to AI, and we intentionally or accidentally sort of cede control of the future to systems that are unbelievably more capable than we are, and we can't get it back," Altman said. "And the other is like, one person or company or country gets way too powerful relative to everybody else."

The first scenario isn't a robot uprising in the Hollywood sense. It's a slow handover: humans gradually let AI make more decisions because it's convenient, or because the systems are better at it, until taking back control is no longer realistic. The second is more familiar territory — a single entity with a decisive AI advantage that everyone else has to live with.

Why an optimist is issuing the warning

Altman has spent most of his public appearances arguing that AI will transform the world for the better. That makes his doomsday scenarios more interesting, not less.

As the CEO of OpenAI, Altman is arguably the most recognizable face of the AI boom. When he says the technology could go "really wrong," it carries more weight than when a critic does — and it's harder to dismiss as marketing.

His comments also track with a shift in the industry's own rhetoric. AI systems have moved from answering questions to taking actions: agents can browse websites, fill out forms and complete multi-step tasks with less human supervision. The more responsibility those systems get, the more the loss-of-control scenario stops sounding speculative.

Altman isn't alone in raising the alarm. Geoffrey Hinton, often called the godfather of AI, has gone further, saying we might have been better off never building the technology at all. The difference is that Altman is still building it.

The concentration problem is already visible

The second scenario — one company or country getting too powerful — is easier to spot in today's AI race than the first.

Frontier AI development is concentrated in a handful of labs, mostly in the US and China, backed by some of the largest balance sheets in history. Compute, data and talent are not evenly distributed, and the gap between the leading labs and everyone else is widening, not closing.

That's why regulators in the US and EU have focused on competition as much as safety. The concern isn't only that AI might escape human control; it's that control might never be shared in the first place.

What it means for you

Altman's warnings don't come with a product launch or a price tag. For readers in the US, UK and EU, the practical impact is policy: both scenarios are already part of the debate around AI regulation, including the EU AI Act and ongoing US hearings on frontier model oversight.

If you use ChatGPT or other AI tools daily, the loss-of-control scenario is less about a sudden takeover and more about how much decision-making you hand over without noticing — from scheduling to shopping to drafting work emails.

Altman's dual warning is a reminder that AI risk isn't one problem but at least two — and they pull in opposite directions. Guarding against a loss of control means slowing down and building oversight; guarding against concentration means spreading access and power. Doing both at once is the hard part, and nobody has a clear plan for it yet.