🤖 Decoy LLM API Proxy

Fake OpenAI/Anthropic API keys routing to monitored proxy.

→ Explore in ATT&CK matrix (1 mappings) Deployment detail

What it is

Plant fake OPENAI_API_KEY, ANTHROPIC_API_KEY, or GEMINI_API_KEY values in environment variables, .env files, and CI/CD secrets. These keys route to a DOAR-controlled proxy that mimics the real LLM API, logging every prompt the attacker sends (revealing their intent, tooling, and objectives) while returning delayed or hallucinatory responses.

Implementation requirements

Reverse proxy mimicking OpenAI/Anthropic/Google API schemas (JSON-compatible)
Valid-looking API key format (e.g., sk-proj-XXXX... for OpenAI)
Response generation: Either cached generic responses or a cheap local LLM
Full request logging: capture prompt, model selection, temperature, tokens
Latency simulation to avoid detection via response-time analysis
Rollback: Remove the planted keys from every location, then shut down the proxy.
Upkeep: Keep the proxy's responses in step with the real API's current models and formats. Monthly proof-of-life request.

Technologies

AI/LLMCI/CD

ATT&CK coverage 1

Technique	Tactic	Status	Confidence	Tier
T1552.001 Credentials In Files	Credential Access	confirmed	strong	T2
↳ Detection point: the decoy fires at a later step of the attack
↳ Evidence (5 sources) CloudKeys in the Air: Tracking Malicious Operations of Exposed IAM Keys — AWS keys planted on GitHub were harvested and used by attackers within minutes. 2023 Honeypotting in the Cloud Report — AWS keys planted in GitHub, S3, ECR, EBS and other resources were found and used; GitHub keys within 2 minutes. GitGuardian Honeytoken — core concepts — Production honeytokens placed in code repositories, CI/CD, Docker images, Jira, Confluence and Slack alert when used. AWS API Keys token — Canarytokens documentation — A fake AWS key alerts on any use through the AWS API. MITRE ATT&CK Enterprise v20.0 — ATT&CK v20 documents T1552.001 Credentials In Files as an adversary behaviour.

Evidence

LLM Agent Honeypot: Monitoring AI Hacking Agents in the Wild — Reworr; Volkov, D., arXiv:2410.13919 2024 peer-reviewed
LLM Honeypot: Leveraging Large Language Models as Advanced Interactive Honeypot Systems — Otal, H.T.; Canbaz, M.A., arXiv:2409.08234 2024 peer-reviewed
OWASP Top 10 for LLM Applications 2025 — OWASP 2025 standard
ATLAS: Adversarial Threat Landscape for AI Systems — MITRE standard
MITRE ATT&CK Enterprise v20.0 — MITRE 2026 standard

Research basis: Extension of LLM Honeypot research (Volkov & Reworr, 2024; Otal & Canbaz, 2024).