Most recent
navigate open esc close Corpus index built 2026-06-07 23:58 UTC

← All reference

Reference

11. Credential Leaks & Misconfigs

Source: https://github.com/nuclide-research/AI-LLM-Infrastructure-OSINT/blob/main/shodan/queries/11-credential-leaks

Section verified: April 2026

The high-impact category. Exposed environment files and provider keys collapse the attack chain from “exposed service” to “full provider account takeover” in a single step.

Exposed API Keys

Shodan QueryNotes
"OPENAI_API_KEY" port:8000Env var leaked in banner/error page
"sk-" "openai" port:3000OpenAI classic key format
"sk-proj" port:3000 OR port:8000OpenAI project keys (newer format)
"ANTHROPIC_API_KEY"Env-var name form
"sk-ant-" -site:anthropic.comRaw Anthropic key prefix, catches leaks where the env var name was stripped
"sk-ant-api03-"Anthropic v3 key format, narrower regex
http.html:"MISTRAL_API_KEY"Mistral env var leaked in page body
http.html:"DEEPSEEK_API_KEY"DeepSeek env var, frequent in 2026 self-hosted stacks
"gsk_" OR "GROQ_API_KEY" port:8000Groq API keys (fast-growing in 2026)
"AIzaSy" OR "GOOGLE_API_KEY" "generativelanguage"Gemini / Google AI keys
"hf_" OR "HF_TOKEN" port:8080Hugging Face token
".env" "API_KEY" port:8000Exposed .env file
http.html:".env" ("OPENAI" OR "ANTHROPIC" OR "GROQ").env files leaking multiple providers
".claude/settings.json"Claude Code project config, reveals enabled MCP servers, hooks, allowed tools; pivots to §10

Reporting note: Keys surfaced via Shodan are already exposed to the index and to anyone who paid for a Shodan subscription. Rotation is the fix; reporting to the key owner is the disclosure. Do not test the key against the provider API, that crosses the line from passive recon into credentialed access.