1. Five governments just put a clock on AI-powered hacking: months, not years.
The agencies built to hedge stopped hedging.
Key fact: Close 3 deferred items from the advisory's list: pull needless internet-facing machines, MFA every admin account, confirm a read incident plan.
cyberscoop.com · cryptobriefing.com · primary source
2. Congress set a vote for Wednesday on who pays for data centers' electricity: the tech giants, or you.
The data-center fight reached the one room that can write the rule.
Key fact: Check your state PUC for a pending data-center cost-allocation case — that's where your bill is actually decided.
eenews.net · whitehouse.gov · primary source
3. China is building AI to flag citizens as political risks before they act — and it's already exporting the toolkit.
Predictive policing leaves the demo stage: the model decides who's a threat before anything happens.
Key fact: If you build or buy ML: the leaked stack is behavioral risk-scoring from telecom + location + social signals, with 'detect harmful information' classifiers as the dual-use core — the same primitive ships inside 'safety' tooling everywhere. Treat pre-act risk scoring as a red line in procurement.
nytimes.com · oodaloop.com · primary source
4. An AI agent out-diagnosed a panel of six doctors in a peer-reviewed study — in a sandbox, not a hospital.
The good news and the asterisk arrive together.
Key fact: In health care: pilot on structured, data-rich cases with a human signing every call; hard stop before real patients outside a trial.
nature.com · medicalxpress.com · primary source