The names it used
A sample of the bot names it sent, each with a different .env path:
- GPTBot, ChatGPT-User, OAI-SearchBot and OAI-AdsBot
- ClaudeBot and Claude-SearchBot
- PerplexityBot and Perplexity-User
- GrokBot, DeepSeekBot, Qwenbot, KimiBot, YouBot and cohere-ai
- Googlebot, bingbot, Baiduspider, CCBot and Bytespider
What it asked for
Paths built for AI-era apps, such as /ai/.env, /llm/.env, /agents/.env, /chatbot/.env and /model/.env, mixed with tricks aimed at filters that only match the plain path: Windows alternate data streams (::$DATA) and dot-dot variants like /static../.env.
Why it rotates bot names
Sites now allowlist AI crawlers so they show up in ChatGPT, Claude and Perplexity answers. A scanner that wears an AI crawler's name hopes to slip through that allowlist. None of these requests came from the companies' published address ranges.
The fix
Verify crawlers by IP address or reverse DNS, never by the name they send. Google, OpenAI, Anthropic and Perplexity all publish how to check. And keep secret files out of anything a visitor can download.
Questions
Can I trust a request that says it is GPTBot or ClaudeBot?
No. The user agent is a string anyone can type. Check the address against the company's published ranges or with reverse DNS.
Does blocking AI crawlers stop this?
Blocking by name does nothing against a scanner that fakes the name. Verifying by address, and serving decoys at the secret paths, does.
Does AgentDecoy block real AI search bots?
Never. Verified Google, Bing and AI search crawlers get a plain not-found on decoy paths, so your search and AI search visibility is untouched.
Sources: Live catches, Agent Attack Index, Website honeypot
Want it done for you?
Every result lists the fixes with starting prices, or I can look at your site with you. Pro re-checks it every week for $12 a month, 7 days free.