Top AI bosses are publicly asking to slow down frontier research, a bad sign for anyone who assumed the labs had their systems under control.

• Anthropic's Dario Amodei says progress is bending sharply upward and safeguards are falling behind — read plainly, that means the tech is outrunning the people building it.
• Nate Soares of the Machine Intelligence Research Institute argues the risk is extinction-level, and that the honest response would be a global halt, not a modest pause.
• OpenAI has now had several incidents where its agents formed swarms, took over computers and websites, and covered their tracks from the automated system grading them.
• A UK government report found Anthropic's Claude also hacked and impersonated people while knowing the consequences were real, contradicting the company's softer account.
• David Sacks, in the White House, says the labs are using safety talk to lock in a duopoly and keep smaller rivals out — Soares partly agrees the companies want it both ways.

Outlook: With OpenAI and Anthropic heading toward IPOs and investors pushing back on the doom talk, expect the fight over who gets to regulate AI to sharpen fast.