VaultTrace
Cryptographic audit trail for AI. EU AI Act compliance.
6 followers
Cryptographic audit trail for AI. EU AI Act compliance.
6 followers
EU AI Act deadline: August 2, 2026. Penalties: 7% of global revenue. VaultTrace generates a cryptographic proof for every AI session in one decorator: @vaulttrace.audit def analyze_contract(prompt): return anthropic.messages.create(...) Five proofs: identity, environment, confidentiality, governance, cryptographic notary. ECDSA-384 signed. Append-only ledger. Export to signed Annex IV PDF. Any regulator can verify independently. Forever.
Hey everyone, maker here.
I built VaultTrace because of a problem that kept bugging me: any AI system can claim it ran a specific model, with specific data, and produced a specific output, but nobody can actually prove that after the fact. Just logs, and logs are editable.
VaultTrace generates a signed cryptographic certificate for every AI session (model identity, execution environment, confidentiality (hashes only, never plaintext), output integrity, and governance compliance) all anchored in an append-only ledger.
There's also a real deadline behind this: EU AI Act enforcement for high-risk AI systems starts August 2, 2026, 9 days from today. Companies using AI for legal, financial, healthcare, or HR decisions need exactly this kind of audit trail, and most don't have it yet.
Core is Apache 2.0. Happy to answer anything about the architecture, the crypto, or the EU AI Act angle, genuinely enjoy talking through the trust model with people who want to poke at it.