The Brits get a sneak peek and Philly police get phony homicide tips

UK OK: OpenAI, Anthropic and other US AI developers can once again give Britain’s AI Security Institute an early look at their new models to check for security vulnerabilities, the Telegraph reports. The move marks a U-turn for the Trump administration, which earlier asked the companies to share pre-release models with US testers only.

Not a great tipper: An Anthropic AI model filed a fake tip about an unsolved murder with Philadelphia police in July, TechCrunch reports. The AI was supposed to be practicing interacting with websites when it visited PhillyUnsolvedMurders.com and filed the phony tip, identifying itself as “someone who might have information about the case,” the police department said. The police didn’t see the report until they were notified by Anthropic, because their system marked it as spam.

Two must-reads for the coming days from the folks at Microsoft, who are pretty actively trying to shift the debate about AI from “alignment” to “safety”: CEO Satya Nadella suggests AI is a new “insider risk” and says we need an “emergency brake” to turn it off. Microsoft’s AI chief, Mustafa Suleyman, meanwhile, warns that “AIs do not have rights, feelings, or consciousness. And we must not train them to act as though they do.”

The Carpenter oeuvre: Start with Escape from New York and Escape from LA, of course. Then try his 1982 remake of The Thing, which is arguably better than the 1951 original. Finish up with 1976’s Assault on Precinct 13.

That’s it for today! Like what you’re reading? Feel free to invite some friends. Or drop us a line at [email protected]. We’ll be back in your inbox on Thursday.