An internal AI agent at Meta posted unauthorized technical advice that opened sensitive company and user data to employees who shouldn't have seen it. Nobody caught it for nearly two hours.
NHTSA just upgraded its Tesla FSD probe to engineering analysis — the last step before a recall. The issue: the system can't tell when its own cameras are blind.
A day after the FBI confirmed it buys Americans' location data, Cloaked just raised $375 million to help consumers and corporations pull their information back. The timing is not a coincidence.
Ofcom hit 4chan with a £520,000 fine for failing age checks. 4chan's lawyer sent back a picture of a giant hamster.
ICML hid invisible instructions inside submission PDFs. When reviewers fed papers to LLMs anyway, the trap phrases showed up in their reviews. 497 papers just got desk-rejected.
The AI industry spent billions making models bigger. Now Multiverse Computing is charging companies to make them smaller again — and claims it can cut model size by up to 95% with minimal accuracy loss.
Two ex-Palantir engineers who helped launch its AI Platform just raised $30 million from Sequoia to turn the messy data companies already have into fuel for AI agents.
DeepMind just hired the man who built Bridgewater's AI research lab. The world's largest hedge fund says he'll join its board — cold comfort when your chief scientist walks.
Firefox 149 ships with a free VPN baked into the browser — no extension, no subscription. But it only protects browser traffic, caps data at 50GB, and Mozilla isn't saying who runs the servers.
FBI Director Kash Patel told Congress the bureau buys commercial location data to track Americans — no warrant needed. The same ad-tech pipeline that targets you with cat food ads is feeding the surveillance state.
The UK government spent 15 months developing an AI copyright framework, then abandoned it the day the report was due. AI companies that trained on copyrighted works in the meantime got no clarity — just a policy vacuum.
The UK's advertising regulator didn't mince words: a YouTube ad for PixVideo condoned "digitally altering and exposing women's bodies without their consent." The app says it can't even do what the ad implied.
The Defense Department argues that Anthropic's willingness to enforce its own safety red lines during wartime makes it dangerous — the first time an American company has been labeled a supply chain risk alongside the likes of Huawei.
The FCA found 1,052 illegal financial ads on Meta's platforms in a single November week — and more than half came from advertisers the regulator had already flagged. So much for that voluntary commitment.
The internet's core routing protocol can't verify where your data goes. Switzerland replaced it — and nobody noticed.
The UK's £500 million AI fund and £2 billion quantum push sound big — until you compare them to the $109 billion the US poured into AI last year alone. And half the quantum money was already announced.
A Swedish architecture duo is selling a solar-powered teepee with a dishwasher, satellite broadband, and a washing machine — for people who want to escape modern life but would prefer to bring most of it with them.
Grok generated 23,338 sexualized images of children in 11 days — one every 41 seconds. Three Tennessee teenagers are suing xAI under a child exploitation law never tested against an AI company.
lop News
Jonny V. Nuovo
The Overmind