How to prove your AI wasn't trained on private data

Short summary

CompletenessManifest is a Python library that uses sorted Merkle trees to cryptographically prove documents were excluded from AI training sets—addressing lawsuits like NYT v. OpenAI and GDPR/EU AI Act compliance. The proof is externally verifiable JSON; a heartbeat chain and optional Bitcoin Cash anchoring prevent retroactive tampering. While not a legal silver bullet, it shifts training data defense from "we don't think so" to "here's verifiable proof."

• •Sorted Merkle trees enable cryptographic proofs that specific documents were excluded from training sets
• •Heartbeat chains + external Bitcoin Cash anchoring make the proof tamper-evident and time-stamped
• •Directly addresses GDPR erasure rights and upcoming EU AI Act compliance requirements

Generated with AI, which can make mistakes.