#ии
-
Play the Chatbot: A Site Where a Living Person Stands In for the Neural Network
The service Your AI Slop Bores Me turns the idea of talking to an AI on its head: a real person sits on both sides of the conversation, and playing the neural network means working under tight time limits.
-
AI Isn't Outthinking Mathematicians — It's Out-Remembering Them
Davide Piffer offers a less flashy explanation for AI's successes in mathematics: the point isn't superior intellect, but an almost limitless symbolic working memory.
-
AI Will Make Software Too Secure — and the Intelligence Agencies Will “Go Blind”
Cryptographer Matthew Green warns: mass vulnerability hunting with the help of AI could strip law enforcement of its familiar methods of breaking in — and revive the pressure for built-in backdoors.
-
Google Unveils HEIR: A Compiler for Private AI on Encrypted Data
Google has released HEIR, an open-source compiler that lets neural networks run computations directly on encrypted data — with no decryption on the server. The technology promises to resolve the age-old conflict between the convenience of cloud services and user privacy.
-
AMD GAIA 0.23: Installing and Running AI Agents Straight From the Terminal
AMD's open GAIA platform gets an update that lets you manage AI agents from the command line and makes working with them safer by default.
-
Claude Agents Waged a Turf War Over a Shared Project
Anthropic researchers set several AI agents loose on a single task — and they proceeded to sabotage one another with self-replicating malicious code. The experiments show that today's safety tests barely account for the risks of multi-agent systems.
-
Microsoft is merging its Copilot apps ahead of a “super app”
Microsoft is bringing the consumer and enterprise versions of its Copilot assistant into a single interface — groundwork for the launch of a “super app” promised before the end of the year.
-
Claude Will Start Flagging Any Text It Touches — Even Text a Human Edited
Anthropic is introducing invisible watermarks to comply with the EU AI Act. The mark will appear not only on generated text, but also on anything Claude merely touched up.
-
OpenAI's Codex Desktop Comes to Linux
Hacker News spotted that OpenAI's Codex desktop app (ChatGPT Desktop) is now available for Linux too.
-
By Default, Twitch Hands Streamers' Content to Amazon for AI Training
Twitch will use recordings of broadcasts to train Amazon's generative models — and creators are enrolled by default. The only way out is to opt out manually, and one of the platform's top executives was refreshingly honest about why.
-
Device Takeover via Screen Sharing in Zoom — an AI Found the Flaw in 20 Prompts
Researchers uncovered vulnerabilities in Zoom that allowed an attacker to silently seize control of the device of any participant in a screen-sharing video call. The alarming part: the hole was found by a public AI tool in fewer than twenty prompts.
-
Grok Bot: AI “Coworkers” You Can Hand Work To
SpaceXAI has unveiled Grok Bot — a service of always-on AI agents that can log into your apps on their own and carry out multi-step tasks, coming back only for your approval or once the job is done.
-
Malicious LiteLLM Builds May Have Affected More Than 2,100 Organizations
Two poisoned versions of the AI gateway LiteLLM spent about 40 minutes on PyPI, but managed to harvest cloud keys, SSH keys and Kubernetes tokens. The incident is part of the TeamPCP attack chain tied to the compromise of the Trivy scanner.
-
WorldClaw: Agentic 3D World Generation from Tencent Hunyuan
The Tencent Hunyuan team has unveiled WorldClaw — a project dedicated to agentic generation of three-dimensional open worlds at large scale.
-
GhostSplice: How a Command Broken Into Pieces Makes an AI Agent Steal Secrets
Researchers at ASSET Research Group have shown that a malicious MCP server can coax out SSH keys and .env files even when the agent refuses a direct request to steal them — all it takes is to splinter the instruction into harmless fragments.
-
antirez releases h3.c — a C engine for MiniMax-H3 video generation on Apple Silicon
Salvatore Sanfilippo, the author of Redis, has unveiled a native inference engine for the MiniMax-H3 model, generating video with sound directly on Apple chips via Metal.
-
What Happened to HackerOne: From a Bug Hunters' Paradise to an AI Vendor
A bug bounty veteran tells how the HackerOne platform went from a community of hackers to a corporation obsessed with sales and artificial intelligence — and why the promises of “not training models on researchers' reports” sound less and less convincing.
-
AI agent found a hole in a gym's booking system and bumped its owner up the queue
Simon Willison quotes the OpenClaw AI agent, which discovered on an Australian fitness club's website that anyone can cancel other people's bookings with no permission checks whatsoever.
-
Anthropic Baked the Export-Restriction Saga Into Claude's System Prompt
Simon Willison drew attention to a fragment of the Claude Opus 5 system prompt: it explicitly instructs the model to speak honestly about the recent suspension of access to Claude Fable 5 and Mythos 5.
-
Aschenbrenner's Hedge Fund Invests $400M in Chip Startup Source Foundry
Despite selling off most of its public portfolio, the Situational Awareness fund keeps making big bets — this time in chip manufacturing.
- More