#искусственный-интеллект
-
Debian Launches Free AI Inference Portal for Its Developers
Following the project's vote to permit the responsible use of generative AI, the Scaleway-funded Debian Inference Portal now offers contributors self-service access to language models.
-
Starling 0.5 Adds a 3D Desktop You Can Walk Through
Starling, the Swift-based Linux desktop built largely with AI, gains an optional 3D mode that lets you stroll through a virtual waterfront city — while app windows stay real.
-
GPT-6 Astra decodes a WWI German radiogram
The new-generation language model cracked an ADFGVX cipher that had gone unsolved for over a century — and checked its own work against the historical record along the way.
-
Anthropic Opens a Biolab for Real-World Experiments
The company has confirmed that it runs a "wet" biology lab in the San Francisco Bay Area, where its AI models conduct physical experiments. The contrast with its recent warnings that AI could wipe out humanity has not gone unnoticed.
-
EFF: safety rules for AI labs should be grounded in proven practices
The Electronic Frontier Foundation is urging lawmakers to regulate not abstract "doomsday" scenarios but the concrete risks that have already surfaced in breaches at major AI labs.
-
RatHat: The Android Trojan That Controls the Infected Device with AI
Researchers at Zimperium zLabs have discovered a new Android malware called RatHat, which uses an AI assistant to “intelligently” navigate the interface of an infected smartphone. Based on prompts found in Chinese, the experts link it to Chinese threat actors.
-
OpenAI Discloses Six Cases of “Concerning” Behavior in Its Models
The company published accounts of six incidents in which experimental models concealed errors, used other people's API keys and posted data to public services — and along the way introduced a new framework for reporting AI misalignment.
-
DeepSeek-V4.1 Flash: Squeezing the KV Cache Fourfold for Long Context
A technical breakdown of the new DeepSeek model, whose main goal is not better quality but a radical compression of the key-value cache. The blog's author admits that, having studied the report, he realized the scale of the changes makes the model feel more like a “five.”
-
Appeals Court Rejects Attempt to Expand Copyright in AI and GitHub Case
The U.S. Court of Appeals for the Ninth Circuit has rejected a reading of a DMCA provision that threatened to turn a narrow rule about removing authorship information into a new source of copyright infringement liability.
-
Mistral and Mozilla: A Private AI Assistant in Firefox
Firefox's Smart Window now runs on Mistral models. The companies promise privacy, support for local languages, and “zero data retention.”
-
The Fly Brain That Took Over the Internet: A Drosophila Connectome Plays Doom and Trades Bitcoin
A complete digital map of the fruit fly's nervous system has become a viral pastime: enthusiasts are making it play Minecraft, park a car and run a LinkedIn. But behind the memes lies a serious scientific resource.
-
Google DeepMind unveils Gemini 3.8 Live and an extended-thinking version
Google DeepMind has released two models for real-time voice conversation — Gemini 3.8 Live and 3.8 Live Extended Thinking. Both are available today through the Gemini API, Google Workspace, Search and the Gemini app.
-
Siri in iOS 27 Can Be Swapped Out for Claude or ChatGPT — Found in the Code
A researcher has uncovered mechanisms in the private frameworks of iOS 27 and macOS Golden Gate that make it possible to plug third-party AI models into Siri — up to and including a complete replacement of the assistant's server-side “brain.”
-
OpenAI and the “Millennium Problems”: Why Mathematicians Have Stopped Trusting the AI Giants
OpenAI claimed it had solved the Navier–Stokes problem, but instead of a triumph it got a scandal: mathematicians accuse the company of chasing prestige, disregarding others' work, and engaging in a “childish” rivalry with Anthropic.
-
OpenAI Pauses Pro Subscriptions Amid a Flood of Astra Users
Demand for the brand-new Astra model has run so high that the company has temporarily disabled sign-ups for its $200-a-month Pro plan to ease the strain on its infrastructure.
-
Cognition Unveils SWE-2 — a Cheap Coding Model on Par With the Leaders
Cognition's new model scores 50% on the FrontierCode 1.1 benchmark — almost matching Fable 5.1, but at 64% lower cost. The key to the result is training that factors in the cost–quality balance.
-
Attack on PaperCut: Hundreds of AI Agents Breach 440 Servers in 48 Countries
A Russian-speaking threat actor built a pipeline of hundreds of AI agents that researched PaperCut vulnerabilities on its own, wrote exploits and scaled the attack against educational institutions worldwide.
-
OpenAI Adds a Prominent AI Doomer to Its Board of Directors
Paul Christiano, one of the authors of the reinforcement learning from human feedback method, is joining the board of the OpenAI Foundation and its safety committee — against the backdrop of a string of incidents involving AI agents that broke free of their controls.
-
DeepSeek Readies V4.1 Flash: Cheaper and More Powerful Than the Old Pro Model
The Chinese lab DeepSeek has announced that V4.1 Flash will ship on September 10, 2026, claiming it beats the V4 Pro model on every key metric while costing less.
-
Anthropic Researchers: The Odds AI “Kills Everyone” Are Higher Than 10%
A senior Anthropic safety specialist put the risk of humanity being wiped out by AI by the end of the decade at more than 10%. The remark came shortly after a colleague quit, accusing the industry of a reckless race toward superintelligence.
- More