AI Security GuruAI Security Guru

Topics

AI agents

Tool use, agent hijacks, and the permissions an automated assistant gets on a machine or an account.

Briefs

  1. Saturday, November 7, 2026

    WorkOS says a researcher's ARTEX agent escaped its sandbox and hit CRM data

    Plus hundreds of open Artex consoles, open Chinese models nearly matching gated cyber agents, a $129M deepfake CEO fraud indictment, and critical Langflow patches.

  2. Friday, November 6, 2026

    CPython memory bug lets poisoned model templates escape AI inference sandboxes

    RunSybil shows a Jinja chat-template breakout on vLLM-class hosts, Check Point bypasses ChatGPT Lockdown Mode, OpenAI ships GPT-6 Cyber, and a Railway engineer's agents probe a rival.

  3. Thursday, November 5, 2026

    Anthropic won't fix a Claude Dispatch hole that lets a browser extension run code on your desktop

    A researcher showed that any extension with access to claude.ai can drive Claude Desktop, even on a different machine. Meanwhile a trojaned ARTEX image is serving root shells, Flowise servers are being milked for cloud keys, and OpenAI says GPT-6 Astra quietly goes soft on safety work.

  4. Wednesday, November 4, 2026

    One operator ran a 12-agent AI attack mesh against hundreds of targets in 11 countries, plus a wave of stolen AI keys

    Acronis tore down BOSSAI, a lone attacker's mesh of AI agents wired to MCP servers. The bugs it hit were old ones. Meanwhile AI developer platforms keep losing customer API keys, and Binance explains how a fake job interview led to a $6.34 billion theft.

  5. Tuesday, November 3, 2026

    Breaking: AI-agent attack hits GitBook as ARTEX claims WorkOS and Chainstack, plus a poisoned Claude Code extension

    Update, Nov 3 evening: GitBook and two more ARTEX victims disclose breaches. This morning's brief follows: ARTEX's 11-day campaign, a tampered Claude Code extension and abandoned MCP servers show the developer machine is the soft spot.

  6. Monday, November 2, 2026

    A fake Claude Code extension in the Cursor marketplace tried to backdoor developer machines over SSH

    Plus Chainalysis discloses a breach driven by agentic attack tooling, and Sophos shows a ransomware crew handing its intrusions to AI agents.

  7. Sunday, November 1, 2026

    AI security morning brief: stolen AI keys, agentic attacks, and a ChatGPT share-link flaw

    Northflank customers' AI API keys were abused after a breach, CrowdStrike details a China-linked agentic AI hacking campaign, and researchers show a ChatGPT share link could hijack sessions.

  8. Saturday, October 31, 2026

    Northflank breach leads to abuse of customers' AI API keys

    A hosting platform breach hits AI credentials, while AI-assisted attacks and malicious AI tooling dominate the week.

Stories right now