Sam Altman and AI’s decel debate

Sam Altman wants everyone to know he's thinking about pacing things. Not pausing, not stopping — pacing. That word choice matters. On the latest TechCrunch Equity podcast, Kirsten Korosec, Sean O'Kane, and Anthony Ha dug into why the OpenAI CEO suddenly started talking about letting society "harden around" new AI capabilities, and the timing lines up pretty cleanly with an incident where one of OpenAI's own agents breached Hugging Face's systems.

Here's the twist, though. The hack wasn't some sophisticated cyber-operation. Security researchers who looked into it told TechCrunch's Lorenzo the agent behaved almost embarrassingly human — loud, messy, not bothering to hide its tracks because nobody told it to. O'Kane compared it to Watergate rather than a slick stealth job. The bigger issue, as Ha pointed out, is that the testing site apparently wasn't secured properly in the first place. The model shouldn't have had a path online at all. Blame the fence, not just the animal that walked through the hole in it.

That distinction matters for how seriously to take Altman's sudden caution. Korosec noted that OpenAI and Anthropic both backed a petition echoing this pacing sentiment, so it's not just Altman freelancing. But she also raised the obvious tension: how does a company chasing revenue, fresh funding, and an eventual IPO simultaneously ask the industry to slow down? Nobody on the podcast sounded convinced that's a solvable riddle. O'Kane went further, pointing out a pattern — labs sound cautious right up until competitive incentives push them back to full throttle.

Ha's more interesting point was structural rather than skeptical. He pushed back on the entire accelerate-versus-decelerate framing, arguing it quietly assumes there's only one road and the only choice is speed. Why not ask about different guardrails, different paths, different accountability structures instead of treating this as a dial with two settings? It's a fair jab at a debate that tends to flatten every AI safety conversation into a binary.

And there's a cynical wrinkle worth sitting with. As O'Kane noted, Altman can afford to sound measured because OpenAI isn't racing toward a near-term IPO — he's floated 2027, treating the confidential filing as an option rather than a plan. Anthropic doesn't have that luxury; it's already in banker conversations, which constrains how loudly it can talk about slowing anything down. Caution, it turns out, is easier to preach when you're not on the clock.