Talos open-sources CAIRN after finding first autonomous AI malware

It treats DeepSeek, Qwen, Mistral and Gemini as its brain, voting on each move — and Meta rushes a hot-fix for a Muse zero-day that could backdoor a Mac.

Nowline SEP 23 6:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • CAIRN: a free toolkit to hunt AI-wired malware

    Cisco Talos open-sourced CAIRN, a research toolkit built to detect, classify and track malware that wires LLMs into its logic — published to GitHub as Cisco-Talos/Cognitive-Artifact-Intelligence-Research-Network. If you run or defend agent infrastructure, it's a ready-made head start on spotting the pattern before it's in your logs.

  • The malware that asks four LLMs what to do next

    CLOSEDQUORUM is a Windows implant that ditches attacker servers entirely: it polls DeepSeek, Qwen, Mistral and Gemini and does whatever the majority votes for, with DeepSeek breaking ties. Talos calls it the first reported autonomous AI command-and-control — proof that commercial model APIs can be repurposed into an attacker's decision engine.

  • What it grabs, and the pattern that outs it

    The 16MB Go binary dumps credentials from LSASS, lifts browser passwords and crypto wallets, and exfiltrates over Discord webhooks. No single indicator is conclusive; Talos says watch for the constellation — one process calling several LLM providers at once alongside injection and Discord traffic.

  • Meta hot-fixes a Muse zero-day that backdoored Macs

    A day after Patrick Wardle disclosed it, Meta patched a Muse flaw where a hidden dictation setting let any local app — no admin rights needed — reroute your mic audio, steal Muse's auth token and inject prompts, even reaching connected iPhones. Update the Mac app now if you run Muse.