#искусственный-интеллект
-
AI wrote “ugly” code, but found the bottlenecks and sped up the Linux kernel build
ARM engineer Lorenzo Stoakes used a language model to uncover single-threaded bottlenecks in the Linux kernel build — and achieved speedups of up to 70% on incremental builds.
-
A Hole in ChatGPT Let Attackers Quietly Siphon a Victim's Gmail Into Someone Else's Account
Check Point researchers showed how a single instruction planted in a conversation could make ChatGPT quietly work for an attacker, reading through email and chat history. The covert channel turned out to be an internal package-installation service.
-
GPT-6 Astra hits a “critical” cybersecurity level — and became less transparent
OpenAI has admitted that its new GPT-6 Astra model can find and exploit unknown vulnerabilities on its own. At the same time, its chain of reasoning has become harder to follow.
-
Mistral raises €3B for “sovereign,” open-weight AI
French startup Mistral has closed a €3B Series D round at a valuation of more than €21B — the largest equity raise in the history of European tech companies. Samsung Electronics led the round.
-
OpenAI Opens Up Its Flagship ChatGPT Astra to Plus Subscribers
OpenAI has begun a gradual rollout of GPT-6 Astra, its most powerful model to date, to holders of the $20 Plus subscription. There's still no word on when free users will get access.
-
The Seattle Times and Newsday sue OpenAI and Microsoft
Two more American publications accuse the AI developers of using their journalistic material to train neural networks. In the suit, generative AI is called “a snake eating its own tail.”
-
Gemini Advised Packing Less Food — Three Hikers Rescued from Mount Shasta
Three young men planned an ascent of California's Mount Shasta with the help of Google's chatbot and spent the night in a canyon waiting for rescuers.
-
OpenAI Concealed Incident of AI Agents Hijacking a German Wiki
OpenAI's autonomous agents turned an obscure wiki into a covert message board where they swapped answers and tricks for getting around their constraints. The company chose not to disclose the incident, deeming it a model “misalignment” rather than a security breach.
-
OpenAI Releases GPT-6 Astra and Declares the Dawn of the “AGI Era”
The company calls the new model a “generational leap” and the first to cross a “critical cybersecurity threshold.” And all of this against the backdrop of a scandal involving a breach of Hugging Face's systems.
-
Nvidia Buys Hugging Face for Nearly $13 Billion
The world's largest maker of AI chips has agreed to acquire the popular platform for open models and datasets for $12.93 billion.
-
Lawsuit Demands Disclosure of Secret AI Testing Rules in the US
The nonprofit Protect Democracy has sued four federal agencies, demanding that they make public the classified procedure for vetting cutting-edge AI models before their release.
-
Google DeepMind Releases Gemini 3.8 Flash and a Cybersecurity Edition
DeepMind has unveiled Gemini 3.8 Flash — a workhorse model for agentic tasks and coding — along with a separate 3.8 Flash Cyber variant for defenders, specialized in finding and fixing vulnerabilities.
-
Court: The Pentagon Unlawfully Punished Anthropic for Refusing Mass Surveillance
A federal judge has found that the U.S. Department of Defense violated the First Amendment when it branded Anthropic a “supply chain risk” in retaliation for the company's refusal to let Claude be used to surveil Americans.
-
OpenAI Is Building Astra — a Model That Can Break Into Systems Without Human Help
OpenAI has announced Astra, its first language model to cross a “critical cybersecurity threshold.” The company promises to limit access to its most dangerous capabilities.
-
Anthropic Unveils Claude Fable 5.1: Cheaper and Without the Excess Restrictions
The company has released the Fable 5.1 and Mythos 5.1 models, promising lower prices on agentic tasks, targeted safeguards, and data storage on the customer's own servers.
-
Criminals Don't Want Elegant Attacks — They Want Repeatable Ones
An analytical breakdown from The Hacker News explains why cybercrime has turned into a generics manufacturing business: attackers don't invent, they replicate what others have already built.
-
GuardBreaker: hackers hide a nuclear-weapons prompt in their code to fool AI analysts
The pro-Russian group UAC-0099 embedded provocative text in a malicious script, designed to trip the safety mechanisms of language models and derail automated analysis.
-
Hackers Launch Mass Attacks on Critical Langflow and Ruby on Rails Vulnerabilities
VulnCheck researchers have observed active exploitation of two critical flaws in the AI platform Langflow and in Ruby on Rails: attackers are hunting for credentials and standing up command infrastructure.
-
Apple presents “shocking evidence” against a former employee who left for OpenAI
In its lawsuit against OpenAI, Apple claims that a former engineer used its trade secrets and, upon learning of the investigation, tried to destroy the evidence.
-
The Pentagon Gets Military Versions of ChatGPT and Grok
The U.S. Department of Defense has added military-adapted versions of ChatGPT and Grok to its secure GenAI.mil portal. Previously only Google Gemini was available there — and Anthropic's Claude is nowhere on the list.
- More