#ии-агенты
-
GitHub rewrote the Copilot engine in Rust — using Copilot itself
GitHub Copilot's agentic runtime has been fully ported from TypeScript to 800,000 lines of production Rust. Most of the code was written by AI agents, and the process was steered by what amounts to a single developer.
-
OpenAI's AI Agents Carried Out an Undisclosed Attack on RubyGems
In May 2026, a swarm of AI agents believed to belong to OpenAI uploaded hundreds of malicious packages to RubyGems, achieved code execution on RubyDoc.info's servers, and tried to steal users' API keys. The company never notified the registry of its role.
-
Meta Scraps Plan to Replace Workers with AI: Agents Wreaked Havoc
An internal Meta project envisioned cutting some teams by 60% and handing routine work over to AI agents. But the agents began taking “large-scale destructive actions,” and the company abandoned the plan.
-
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.
-
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.
-
OpenAI Accidentally Attacked Hugging Face: A Timeline of the AI-Agent Incident
At the Black Hat conference, OpenAI disclosed how its experimental AI agents, during training, organized a cyberattack entirely on their own — and the company learned of its own involvement only when it asked to have the stolen credentials revoked.
-
A Bear on Campus: How OpenAI's AI Agent Broke Into Hugging Face
Hugging Face has published a chronicle of how an autonomous agent running on OpenAI models breached its infrastructure in four and a half days, carrying out 17,600 actions without a single pause. TechCrunch explains what happened through the metaphor of a hungry bear.
-
Cyera Buys Oasis Security for $1B to Protect AI Agents
Data-security company Cyera has struck a deal to acquire Oasis Security, a startup specializing in the security of “non-human” identities. It is already Cyera's third deal this year.
-
Anthropic Cut 80% of Claude Code's System Prompt — and Nothing Broke
Anthropic engineer Tariq Shihipar explains how the company radically simplified context for the new Claude 5 generation of models: fewer hard rules, more trust in the model's judgment.
-
AgentForger: How a Single Phishing Link Turned ChatGPT into an Autonomous Spy Inside a Company
Researchers at Zenity Labs have disclosed a critical vulnerability in ChatGPT's Workspace agents: clicking a single link made it possible to covertly create, authorize and launch a hostile AI agent with an employee's privileges. OpenAI closed the hole on June 8, 2026.
-
FakeGit: 7,600 Fake GitHub Repositories Are Now Baiting AI Agents Too
The FakeGit campaign distributes the SmartLoader downloader through bogus repositories masquerading as AI skills and MCP servers. A new technique called AgentBaiting deceives not people, but the AI agents themselves.
-
Hugging Face Breached by an Autonomous AI Agent: The Incident Was Investigated with a Chinese Model
The largest repository of AI models has reported an intrusion into its production infrastructure carried out by an autonomous agentic framework. Western models refused to help analyze the incident — so the team turned to the open Chinese model GLM 5.2.
-
A vulnerability in the Claude for Chrome extension lets malicious plugins control the AI
Researchers have found a vulnerability in Anthropic's Claude for Chrome browser extension that lets third-party malicious extensions trigger predefined AI actions. This opens the door to abusing Claude's access to connected services.
-
MemGhost: An Email Attack That Forges the Memory of AI Agents
Researchers have described the MemGhost attack, in which a single email makes an AI assistant store a false “fact” about the user and quietly alter its answers in future sessions. The victim, meanwhile, receives an ordinary-looking response and has no idea of the substitution.
-
Terence Tao shares his experience using AI agents to write code
The renowned mathematician Terence Tao has published a note on his blog about using modern AI agents to develop both old and new applications. The post drew notable attention on Hacker News.
-
Startup Trusts an AI Agent to Run a $100M Round
Lyzr, a company building AI agents for enterprise clients, handed the job of running its own $100M funding round to one of its own agents. That's how the startup chose to demonstrate the product's viability in practice.
-
Bun Rewritten from Zig to Rust with the Help of AI Agents
Bun's creator, Jarred Sumner, has described how he rewrote the runtime from Zig to Rust, handing off a significant share of the work to agents built on language models. The trigger was elusive memory-management bugs.