Cryptographic Forensics for AI Coding Agent Sessions

The DEPOSE project proposes a bundle format that makes AI agent session records cryptographically verifiable.

It separates signed event data from generated narrative, then anchors the log with hash chains, Ed25519 signatures, and RFC 3161 timestamps.

A standalone verifier can detect tampering in the log or supporting files, turning session transcripts into audit-ready evidence.

The approach matters because AI coding agents may hold real credentials, and ordinary JSONL logs are easy to alter after the fact.