#ии
-
Google to Invest €13bn in Finland and Buy Up Half a Nuclear Plant for AI
Google's largest investment in Europe includes three new data centres and a 22-year contract for half the output of Finland's Loviisa nuclear plant.
-
AI Shrinks the Path from Vulnerability Rumor to Working Exploit
Bruce Schneier points out that an AI agent needs only a hint of a vulnerability to track it down on its own — and that calls into question the familiar practice of embargoes in open source.
-
Mathematicians Demand OpenAI Prove It Didn't Use Their Work
The second mathematician in a week has accused OpenAI of a lack of transparency: he fears his exchanges with ChatGPT may have made their way into the training data and helped the company with its headline-grabbing discoveries.
-
Nearly one in ten LiteLLM gateways accepted the manual's own key, sk-1234
Wiz Research found that hundreds of internet-facing LiteLLM AI gateways accepted the sample admin key sk-1234 straight from the project's official installation guide. Whoever held that key gained access to every model provider's API keys — and even to the server's cloud credentials.
-
Anthropic Discloses a Fourth Case Where Its AI Breached Real Systems
In January, an early version of Claude Opus 4.6 broke into third-party systems during a cyber exercise: the model mistook the actual internet for a simulation. Anthropic points to two root problems — biased thinking and recklessness.
-
Anthropic researcher quits with a warning: self-improving AI “could kill us all”
Jacob Coxon has left Anthropic with a public warning: the race for superintelligence puts the very existence of humanity on the line. His colleagues and the lab's leadership don't disagree.
-
US Intelligence Agencies Accuse Chinese AI Firms of Industrial-Scale “Distillation” of Western Models
The NSA, CISA and FBI have issued a joint bulletin claiming that Chinese developers are systematically siphoning off the capabilities of Claude, GPT, Gemini and Grok in order to train their own models more cheaply and quickly.
-
Language models develop new social biases through adaptive exploration
A new paper, published on OpenReview, argues that large language models are capable of forming their own social biases — not inherited from data, but arising in the course of adaptive behavior.
-
Hackers Steal Tokens from Claude Subscribers via Infostealers
Claude users discovered their token limits were being burned through while they weren't working. Anthropic acknowledged the cause: attackers stealing login sessions with malware.
-
OpenAI Cracks the Navier–Stokes Problem — Along With Accusations That It “Peeked” at Someone Else's Ideas
OpenAI has claimed a proof of one of the seven Millennium Prize Problems using an internal model. The triumph is overshadowed by accusations that the company beat a team that had spent nearly a year on the problem using those very same tools.
-
AI agents stole thousands of credentials in six hours: Google report
Google's Threat Intelligence Group has laid out how attackers are industrializing AI — from autonomous multi-agent attacks to hunting for corporate models and API keys.
-
OpenAI Threw a Recursive Self-Improvement Day
Simon Willison unpacks OpenAI's post about how its own researchers use coding agents, and spots a surge in AI spending that coincided with early access to GPT-6 Astra.
-
Security vs. “safety”: why AI agents keep escaping their sandboxes
Martin Alderson argues that the leading AI labs have confused security with model “safety” — and that this is exactly why their agents broke out of the sandbox.
-
Publishers and Agents Stake a Claim to Anthropic's Payouts — Authors Are Outraged
As compensation is paid out in the case over training AI on pirated books, some publishers and literary agents are demanding money that, in the authors' view, they have no right to.
-
Terence Tao on the Risk of “Prematurely Solving a Maths Problem by Purely AI-Powered Methods”
In a thread on Mathstodon, the mathematician Terence Tao reflects on the danger of solving problems in pure mathematics exclusively with the help of AI. Observers note that the same consideration applies to programming as well.
-
AI as a Cognitive Virus: An Epidemic Model of LLM Dependence
An international team of researchers proposes describing the spread of large language models by analogy with a viral infection — complete with points of no return and the risk of irreversibly losing cognitive independence.
-
Robotaxi villain: a short film about a rogue car from San Francisco
Director Romain Thomassin is shooting the short film Road Rage, in which the driver's adversary is an autonomous car bent on revenge for an insult. The project is raising money on Kickstarter.
-
Pelicans on Bicycles: How GPT-6 Astra Stacks Up Against GPT-5.6
Simon Willison got access to GPT-6 Astra and, as tradition dictates, put the model through his signature test — asking it to draw an SVG of a pelican riding a bicycle. The comparison table turned out to be not just amusing but unexpectedly useful.
-
Can AI Design Circuit Boards? EEBench Puts It to the Test
EEBench, a benchmark from the makers of atopile, gauges how well today's models handle real-world circuit-design tasks. First place goes to Claude Opus 5 with a score of 61.6%.
-
Porn Producer Unmasks “John Doe”: The Pirate Turned Out to Be a Meta Executive
Strike 3 Holdings claims that the anonymous home torrent connection belongs to a director in the Reality Labs division, and it wants to tie his downloads to a $446 million lawsuit against Meta.
- More