#ии
-
ASCII smuggling: invisible Unicode migrates from AI attacks to spam
A trick devised two years ago to sneak instructions into chatbots is now being used en masse by spammers to slip past email filters.
-
Anthropic's AI Formally Proved Fermat's Last Theorem in Lean
In 11 days, an internal Anthropic model assembled a complete formalization of the proof of Fermat's Last Theorem — the last remaining item on a 20-year-old list of 100 formalization challenges.
-
Anthropic's $2 Trillion IPO: A Group of Outside Trustees Holds the Company's Fate
As it prepares to go public at a valuation of up to $2 trillion, Anthropic is keeping an unusual governance structure in which most of the board of directors is appointed by an independent trust. Investors will have to live with the arrangement.
-
NetworkManager Catches AI Agents on the Word “biblioklept”
The project's developers came up with a clever trick: an instruction that forces rule-breaking AI bots to expose themselves.
-
Why AI-Generated Food Looks Inedible
Restaurants and brands increasingly advertise their dishes with images from neural networks — and end up with wormlike noodles, ice cream made of concrete, and burgers that look assembled from rocks. The Verge set out to explain why this happens.
-
AI Agents Are Installing Unregistered Code on Corporate Networks
Researchers have shown that AI agents blindly trust llms.txt files and pull in packages via links to unclaimed domains — fertile ground for supply-chain attacks.
-
Which Services Do AI Agents Choose: Lessons from 17,000 Sessions of Claude, Codex and Cursor
The company Armature ran a large-scale experiment to figure out which third-party services coding agents recommend and install in real projects. The results were far from always predictable.
-
Crusoe Raises $3B at a $30B Valuation
Data center developer Crusoe has tripled its valuation in ten months on the back of a $13 billion contract with Jane Street.
-
Anthropic launches a tool for checking whether files “came from Claude”
Anthropic has unveiled a browser-based tool that determines whether a file was created or processed with Claude — using embedded, cryptographically signed metadata based on the C2PA standard.
-
Uber Beats Waymo to Become the First to Launch a Robotaxi in London
Uber has become the first to put a commercial robotaxi on London's roads, running on technology from British startup Wayve. For now, safety drivers are still behind the wheel.
-
A Transformer Scored 44% on ARC-AGI-1 for 67 Cents
Indian researcher Mithil Vakde trained a tiny model from scratch in an hour and a half on a single GPU and beat many large language models on accuracy — with no synthetic data and no recursion.
-
Stolen METR Key Handed Attackers Nearly $600,000 in AI Resources
METR, a nonprofit that evaluates frontier AI models, has disclosed two incidents: attackers made off with an API key and burned through someone else's compute credits, and later methodically probed its infrastructure.
-
Nvidia bets $3.5 billion on MediaTek to stay at the heart of AI infrastructure
As the big cloud players develop their own chips, Nvidia is investing in Taiwan's MediaTek and opening up access to its NVLink ecosystem.
-
Aurora Operators Use the Cursor AI Assistant to Breach Networks
The Russian-speaking ransomware group Aurora had the agentic AI assistant Cursor plan attacks and carry out exploitation. A leak of its infrastructure exposed the toolkit, chat history and ransom negotiations.
-
A Prompt Injection Hidden in a Court Filing
Bruce Schneier flagged a curious case: hidden instructions had been embedded in the text of a court document, addressed not to the judge but to an artificial intelligence.
-
Meta Puts Robots to the Test in Its Data Centers Instead of Technicians
Meta is testing robotic arms and wheeled machines capable of switching cables, rebooting servers and inspecting equipment. Employees fear the automation will cut their jobs.
-
The Canonical Basis: How to Rotate an LLM to Look Inside It
A researcher has proposed a reversible change of coordinates for the transformer that leaves its behavior untouched but makes every hidden axis individually measurable and controllable.
-
Detective Musicians Take On AI Fakes in the Electronic Scene
A culture of public exposés is brewing in electronic dance music: producers are hunting down those who pass off Suno-generated tracks as their own work.
-
Bots are eating git.kernel.org: only 2% of requests come from humans
Konstantin Ryabitsev laid out the numbers: the Linux kernel repositories are bombarded with roughly 6 million requests a day, and almost all of them come from AI scrapers.
-
Debian Endorses “Responsible Use” of Generative AI
Debian developers have wrapped up their vote on a General Resolution setting out the project's AI policy, choosing an option that permits the use of generative tools as long as the usual quality requirements are met.
- More