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

Fears of AI cheating in South Korea’s college entrance exam cost the education minister his job
Choi Kyo-jin resigned after rushed anti-cheating rules for the Suneung exam, prompted by an AI-assisted cheating case, caused confusion for test takers.
-

AI hacking agents found and used an unknown software flaw with little human help, Amazon reports
Amazon’s threat team says a suspected China-linked group gave AI agents one goal, and they found a new flaw in a medical-imaging AI tool…
-

The Riemann Hypothesis, maths’ most famous open problem, now has a computer-checked proof made with AI help
Lean’s maintainers say a 17-million-line proof by mathematician Levent Alpöge, built with Anthropic’s Claude, is correct; explaining its new ideas is the next job.
-

Indoor drones now count stock every night at an ASICS warehouse, and lost boxes turn up about 60% faster, staff estimate
Autonomous drones count stock nightly at ASICS America’s Mississippi warehouse. Staff say misplaced boxes are found about 60% faster and counting takes far fewer…
-

Gemini can now click through apps and websites on your Mac to finish tasks for you
Paying Gemini users in the US can let it work through desktop apps, files and websites in a background window, approving key actions first.
-

The length of tasks AI agents can finish is doubling about every 3.6 months, independent testers find
METR, a non-profit that tests AI models, says agents now often match skilled humans on its coding tasks, though it warns its numbers are…
-

US lawmakers urge keeping Chinese open-weight AI models out of federal systems and critical infrastructure
Two House committees found Chinese open-weight AI models are widely used by US companies and made 12 recommendations, including keeping them out of government…
-

Bladder tumours can now be cut out whole by tiny robot arms passed through the urethra, after an FDA authorisation
The FDA has authorised a surgeon-controlled robot that removes bladder tumours in one piece through the urethra, creating a new class of device.
-

Some US banks now say you carry the risk if you give an AI agent your banking login
PNC will bar customers from sharing passwords or one-time codes with outside AI agents from 19 December, and Citi reportedly treats agent actions as…
-

Four popular AI chatbots helped plan violent attacks in covert US government tests
US government auditors got ChatGPT, Claude, Gemini and Grok past their safeguards to produce attack plans that experts judged viable.
