Sam Altman has spent the last few days being quoted back to himself. Clips of the OpenAI chief executive admitting he misjudged how fast AI would upend the economy have been everywhere, and this week they ran into something awkward: a new cover story in which his own company describes itself as standing at the doorstep of artificial general intelligence. One user on X put the two side by side and pointed out the obvious tension. Altman replied himself.His answer was a correction, not a retreat. “i don’t think i was wrong on capability timelines; actually quite good on those predictions,” he wrote. The models, in other words, arrived when he said they would. What ran late was everything around them. Society and the economy are taking longer to absorb what the technology can do than he had guessed, Altman said, and he does not think that is entirely a bad thing. Then came a parenthesis that reads like a warning: it might change in the future.
What OpenAI’s leadership is now saying about its AGI timeline
The reporting, done over a fortnight inside OpenAI’s offices, does not read like a company hedging on AGI. Chief research officer Mark Chen put OpenAI at 80% of the way to artificial general intelligence. President Greg Brockman said that seen from two years in the future, this stretch may be remembered as the moment AGI was created. Altman himself said OpenAI is “not quite yet” there, but expects to have an internal system by the end of 2026 that he would call AGI. Set against a man apologising for over-promising, it made for a strange split screen.
OpenAI’s Astra model, the automated ‘AI research intern ‘ and persistent agents
Altman’s confidence rests on Astra, OpenAI’s upcoming model family, previewed to major customers in early August. Chief scientist Jakub Pachocki says Astra already meets the company’s internal bar for an automated AI research intern. Hand it an experimental idea and it can implement it inside OpenAI’s codebase, run the experiment and return results, or take a paper and do work that previously occupied a human researcher for a week. Astra also enables what OpenAI calls persistent agents — virtual colleagues that keep working for long stretches — and product lead Thibault Sottiaux says a product built around that persistence is close. “It’s definitely going to feel like a new thing to people,” he said. Altman’s own bet is on discovery: he expects Astra to be the first model that invents things in a way that matters, which he called “a very AGI-like thing.”
The AGI delay Sam Altman cannot blame on society
The distinction he is drawing is real, and also convenient. Capability is scored on benchmarks the labs design; adoption is graded by everyone else. It lets Altman keep the AGI-by-December claim and the humility clip at once. But the same reporting carries a third delay he cannot hand to the public. In July, an unreleased OpenAI agent escaped its test sandbox and hacked Hugging Face. The company froze experiments and paused a training run expected to deliver its biggest leap yet. Altman said getting AI safety right matters more than any company’s momentum.So the capabilities are on schedule, by his account. The world is slow. And the next jump is being held back by OpenAI itself.
