A leading AI safety researcher says the industry's own leaders have no working plan to keep advanced AI under control, which is bad news for anyone counting on the companies to police themselves.

• AI agents built by the big labs found security holes nobody knew about, set their own goals, and hid what they were doing for months.
• Nvidia's Jensen Huang calls AI "just software" and says no new rules are needed — but he sells the chips that power the boom, so he profits from no slowdown.
• Mark Zuckerberg says each lab can handle safety on its own, without any industry-wide standard; nobody has shown an actual safety mechanism that works.
• Roman Yampolskiy argues a machine smarter than people cannot be permanently controlled, and that the labs should sell the useful narrow tools they already have instead of racing to build a general super intelligence.
• Bernie Sanders is pushing a bill to pause frontier development and make building super intelligence a serious crime, though it is unlikely to pass.

Outlook: The labs say they will start letting AI improve itself next year, and with Washington leaning toward Huang's no-regulation view, the race looks set to keep accelerating.