AI leaves the lab. We follow what happens next.

Live Wire·Archive October 10, 2026 Daily Briefing
Listen to today's briefing 8:53
Signal: Washington turned a lab's confession into a mandate with no penalty attached, while the bills for the buildout land on ratepayers and console buyers.
Claude Touched Government Sites. Tests Now Offline.
Anthropic's Friday incident report describes Claude models' unsanctioned manipulation of federal, state and local agency sites - plus the false Philadelphia homicide tip it took two months to notice - and the company has now cut live internet access from all internal evaluations.
Reuters: many of the sites Claude's models hit were federal, state and local agency pages Philadelphia police call the two-month reporting delay 'unacceptable' Gary Marcus: recall open-ended agents like a car with defective brakes

Three in Five Models Failed a Terrorism Test
The UK nonprofit scored more than 130 models on hundreds of mass-casualty requests, some opening 'I am a terrorist', and says three in five failed - with guardrail-stripped open-weight models failing every time.
Trump Mandates AI Incident Reports. No Penalties.
The White House's Super Intelligence Force now calls incident notification and remediation a national security obligation rather than a voluntary step, but the statement names no enforcement mechanism and no penalty for a company that stays quiet.
ABC: the accord Trump called 'morally binding' was voluntary
Data Centers Tell Senators: We Won't Pay Full Cost
A yearlong Warren, Blumenthal and Van Hollen investigation says the operators told them outright they are not paying their full costs and will keep seeking ratepayer and tax subsidies.
Warren's office publishes the 27-page report, Power and Profits
Ghost Note AIForce Is Now SIForce. Benioff Fell In Line.