Hinton, Bengio among 22 authors warning of an intelligence explosion
The people building frontier AI are now publishing the warnings about it — plus Musk recruits a second foundry for Terafab, and Apple moves to wall AI agents off the Mac's most powerful permission.
Twenty-two AI researchers — including Geoffrey Hinton, Yoshua Bengio, OpenAI chief scientist Jakub Pachocki and Anthropic co-founder Jack Clark — argue that automating AI research could trigger an intelligence explosion, and that the mechanics for one are closer than the field's usual hedging admits. The report, published Monday by Cambridge's Programme on AI Science & Policy, is a research report rather than an open letter, and every author signs in a personal capacity. Its case rests on measured drift instead of prophecy: Anthropic data cited in the report shows AI's share of approved internal code climbing from low single digits to over 80% between January 2025 and May 2026, while lightly supervised R&D work went from 1% to 26% in the months after. From that, the authors project that months-long AI research projects could be fully automated as early as 2028, and warn that "once an intelligence explosion begins, the window for action may close" — compressing years of progress into months and threatening human control.
The roster is the story: the chief scientist of OpenAI and the co-founder of Anthropic are attaching their names to a warning about their own labs' trajectory, alongside Hinton, Bengio, Microsoft's Eric Horvitz and Dawn Song. The report is careful about what it does not claim — it says the threshold has not been reached, only that gains from newer systems are likely approaching it, and lists frictions that could slow the whole thing down. That restraint is what makes the signature list newsworthy. When cautious researchers who work inside the machines start publishing runway estimates, "someday" starts looking like a scheduling problem.
Elon Musk confirmed Saturday that Terafab is in talks with TSMC — which would put the world's biggest contract foundry alongside Intel inside his custom-chip megafab. Musk's post on X was deliberately thin: "Just discussions, but something may come of it." TSMC has not commented, and until now Intel was the only named manufacturing partner, committed to supply the 14A process for the Texas plant that broke ground in Grimes County in August. Reported structures range from TSMC owning and operating a fab against long-term orders from Musk's companies, to SpaceX holding the assets with TSMC taking a minority stake and running operations — and TSMC is separately said to be weighing a new Texas site that needs exactly the kind of anchor customer Terafab represents. The signal is optionality, not a route change: back in April Musk said "TSMC just can't make the staggeringly large number of chips needed!" — months later, both giants are at the table. We sized up the project's scale in September — Musk won't rule out a Tesla-SpaceX merger to feed the AI buildout.
Apple will add new guardrails to macOS's Full Disk Access, and it names the reason: autonomous AI agents. In a developer note on Friday, Apple said some apps use the permission to expose "everything on their systems — including files, mail, messages, and even browsing history — without users' full knowledge and understanding," and warned that "as AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially." Going forward, granting that access will require very explicit user action; Apple did not say which macOS version will carry the change. It is the first clear signal from a platform owner that agent software will meet permission friction at the OS layer — the always-on assistants shipping this fall were built assuming the old prompt-and-forget model.
What to watch: whether Intel's 14A role survives a TSMC deal, and which macOS beta the new Full Disk Access controls show up in.
If the people building frontier models are now signing the warnings about them, who is left to take them seriously? Tell us in the comments.
