Fake GPTBot and ClaudeBot scanners are hunting your AI keys

Octopus, Gecko and Tortoise: the 3 scanners hunting your AI keys

Three scanners — Octopus, Gecko, and Tortoise — have been probing ten sites from late September to 18 November 2026. Octopus cycles through 32 AI crawler names, rents a fresh Google Cloud machine per target, and tailors its path list to each site's framework. Gecko, a .NET program based in Belgium, now hunts for Claude Code, Codex, Cursor, and Gemini CLI credentials. Who runs them remains unproven.

If GPTBot or ClaudeBot asks your site for /.env, it is not OpenAI or Anthropic.
  1. mormegil

    Whenever I see similar scanning in my server logs, I feel the urge to... provide them with same fake/poisoned credentials to... well, that's the problem. There is no way to hurt those. Usually I report the scan to Google (as they most often come from GCP IPs), but never hear back and it's not like it would stop them (it's some stolen cards or Google accounts, anyway, right?). It's just... it offends me there is no way to fight back.

  2. AMerrit

    Interesting that in some cases they are using the AI Agent crawler names to hide themselves. With the explosion in bot traffic it gets much easier to hide yourself in the crowd.

  3. nerdralph

    Some useful information, though I'd like more details on headers sent like Sec-Fetch.

    For passive blocking, TLSv1.3 filters out some bad bots and scrapers.

    Most bad bots follow 302 redirect, but don't parse html and act on meta refresh.

    Most bad bots don't store cookies, but I've seen a couple that do.

More from this day

2026-11-18