FAQ — for compliance officers & counsel

Plain answers to the questions a risk, legal, or audit function asks first.

Does adopting the Causal Seal make us compliant with the AI Act?

No — and any vendor who tells you otherwise should worry you. Compliance is a property of your whole organization: risk management, data governance, human oversight, documentation. The Causal Seal supplies one specific technical capability the law presupposes but does not define — verifiable, automatic, per-decision traceability (Article 12). It is a strong, demonstrable building block, not a certificate.

What exactly does a seal prove?

That a given AI output is bound, intact and unaltered, to a specific set of declared decision parameters and a timestamp — verifiable by anyone, without access to the model. With a signature, it also proves which system emitted it. It proves binding, not virtue: it fixes the record so it can be audited, exactly as a signed contract fixes terms without making them fair.

Is this a proprietary product we get locked into?

No. The format is an open standard (specification under CC BY 4.0) and the reference verifier is open-source (MIT). Anyone may implement it, verify seals, or build tooling — free of charge and free of license. You can read the whole specification and run the verifier yourself today.

Can we verify a seal without trusting the vendor?

Yes — that is the point. Verification is a public computation: recompute the fingerprint, compare. Our browser verifier does it entirely on your device, sending nothing anywhere. An auditor never has to take the emitter's word for it.

Does it work with our existing models (OpenAI, Anthropic, open-source)?

Yes. The format is model-agnostic: it records the governing decision state around any model, without modifying it. The one prerequisite is a governance layer that produces an explicit decision state to seal.

How is this different from our audit logs, or from C2PA?

Audit logs prove a record wasn't changed — but a faithful log of an unexplained decision is still unexplained. C2PA proves where a media file came from. The Causal Seal proves why an AI output was produced — the decision itself. They are complementary; a seal can even be stored inside your existing audit log.

Who is behind it, and is it stable enough to build on?

The specification is public, versioned, and archived with a permanent DOI (10.5281/zenodo.21431267). A reference implementation runs in production. The current status is v1.0-draft, open for community review — the right moment for your organization to evaluate it and shape domain profiles for your sector.

What would a pilot look like?

Read the Article 12 mapping, run the verifier against sample seals, and review the specification and reference code. For a governed deployment, sealing can be added at the emission point without retraining or changing your model.

Causal Seal — an open standard for decision provenance. This page is informational only and is not legal advice. DOI 10.5281/zenodo.21431267.