How to Detect If Your LLM API Is Fake

Learn how to identify API fraud — model substitution, token inflation, and relay scams.

API fraud is one of the biggest problems in the AI industry. When you pay for premium model access through a third-party provider, how do you know you are actually getting the real model?

Three Types of API Fraud

Model Substitution — The relay claims to serve GPT-4 or Claude, but routes requests to a cheaper model. A system prompt makes the model self-identify as the premium one.

Token Inflation — Reported token counts are 1.5-3x higher than actual usage. Hidden system prompts add 50+ tokens you pay for but cannot see.

Relay Chain Fraud — Data passes through intermediary servers. Each hop adds latency, potential data logging, and injection risk.

How to Detect Fraud

Without an API key, you can check IP/ASN ownership, server headers, architecture fingerprints, and domain verification.

With an API key, behavioral fingerprinting reveals the actual model behind the endpoint. Token audits compare reported vs estimated counts.