Causal Seal

Content provenance proves where an artifact comes from. Log integrity proves a record wasn't altered. The Causal Seal proves what neither does: why a generative AI system produced a specific output — by cryptographically binding the output to the causal parameters that governed its generation. Open format. Model-agnostic. Verifiable by anyone.

🟢 Verify a sealPaste a seal, get a verdict — computed entirely in your browser, nothing sent anywhere. 📐 Read the specificationv1.0 — data model, canonicalization, verification levels, conformance. 💻 Get the codeZero-dependency reference implementation, JSON Schema, computed test vectors. MIT. 📄 Read the paperCausal Seals: Decision Provenance for Governed Generation. 🔌 Drop-in gatewayOpenAI-compatible proxy that seals every response — streaming included, zero code change. ⚡ Install & quickstartpip install, emit & verify in 60 seconds, or any language.

For compliance officers and counsel — the EU AI Act (Articles 12 & 19, enforceable for high-risk systems from 2 August 2026) requires automatic event recording and traceability of AI system functioning, without prescribing a technical form of proof. The Causal Seal supplies that technical capability: one verifiable record per output, covering who answered, with what, under which constraints, seeing what, and when.

Dedicated guidance: What is decision provenance? · EU AI Act mapping · NIST AI RMF · US state regimes · FAQ.

Causal Seal v1.0 · specification text CC BY 4.0 · code MIT · published for community review.
Reference implementation in production: chat.baten.ai (BATEN Technologies, first conformant implementer).