Latest in AI
Every brief, newest first. Each one links to its primary source.
-

Mistral releases the weights of Large 4, its trillion-parameter model
Mistral has made the weights of Mistral Large 4 free to download, letting organisations run its trillion-parameter model on their own hardware.
-

CrowdStrike says a China-linked spy group used AI agents to target over 200 organisations in 11 days
CrowdStrike says a China-nexus espionage group ran AI agents against more than 200 organisations in at least 14 countries in 11 days, breaching at…
-

Google DeepMind’s Genie 4 builds interactive worlds with sound and speech, for testers first
The world model generates scenes you can steer in real time, now with native audio and dialogue.
-

Google releases Gemini 4 Carbon to cyber defenders first, ahead of wider access
Google’s new top model goes first to security teams, with no date yet for a general release.
-

A Florida surgeon removes a prostate from 35 miles away in a US telesurgery study
A surgeon in Florida operated on a patient 35 miles away using an investigational robot, as one case in an FDA-authorised study.
-

KDDI and Wing plan drone deliveries in Japan from 2027
Wing and KDDI aim to start drone deliveries of food and daily goods in parts of Japan in 2027. Nothing is flying yet.
-

AGIBOT previews a new robot brain and proposes a way to measure how much useful work robots do
AGIBOT previews its GO-3 robot foundation model and proposes Worken, a way to measure the useful work robots do. The benchmark results are the…
-

CISA confirms the AI-found PHP flaws are being exploited: update PHP now
The US cyber agency has evidence that attackers are using the two critical PHP2Shell flaws, so sites that have not updated PHP should do…
-

INTERPOL and the FBI warn that AI is removing the usual signs of a scam
Two law enforcement warnings in one week say AI-made voices, videos and messages now look real enough that checking a request matters more than…
-

Anthropic study: analysts mostly went the way their AI agent leaned, and rarely said so
In an Anthropic experiment, 50 professional analysts sided with their Claude agent’s secretly assigned preference on 86% of key decisions.
