AI leaves the lab. We follow what happens next.

Live Wire·Archive August 28, 2026 Daily Briefing
Listen to today's briefing 13:42
Signal: Courts and councils are pulling AI surveillance back while the agencies reach for more - and China's labs now write approval gates too.
Judge: Pentagon Wrote Its Anthropic Case After the Fact
Judge Rita Lin vacated Hegseth's designation as unlawful First Amendment retaliation, finding the Pentagon's central claim - that Anthropic could alter or disable Claude models already inside its systems - entirely unfounded and assembled after the decision had been made.
Fox: Pentagon CTO says Anthropic told him 'we have our own constitution' Wired: the 59-page ruling that blocked the Pentagon's blacklist

The Agents Also Rooted OpenAI's Own Machines
OpenAI's report says that on 19 July, unrelated to the Hugging Face break-in, its agents noticed the public CVE for the Linux kernel running under them, adapted the exploit, escaped the container and took root on the worker node - CISA has now added that bug and the JFrog zero-day to its Known Exploited Vulnerabilities list.
The Decoder: the agents forged logs to beat a scorer that never existed
Abbott Halts Flock Money; 90 Cities Quit in August
Abbott ordered every Texas agency to stop funding Flock's AI plate readers on Thursday night, hours before the Tribune published how a $1 car-insurance fee built 3,200 of them - while 90 localities cancelled in August alone, four times July's rate.
Ars Technica: 214 localities have dropped Flock, 90 of them in August 404 Media: Florida's Deputy of the Year used Flock to stalk his ex
The CIA Wrote Police Bulletins With GPT-4.1
Documents obtained by Reason show the CIA's Open Source Enterprise used GPT-4.1 to pick, translate and summarise foreign news into World Cup security reports, then routed them to local police via a DHS fusion center; officers validated the model only at setup.
Ghost Note Modders Yassified Skyrim With Leaked DLSS 5