Human-in-the-loop evidence
Every organisation using AI says the same sentence: a human checks it. Almost none can prove it — not to a regulator, not to a customer, not in court. The human-in-the-loop record turns that claim into tamper-evident proof.
Not: we review. Rather: here is who reviewed and released what, and when.
For every AI result that leaves your organisation: who released it, when, which result — via a content hash that pins the released text exactly — and in which product, on which object (case, article, e-mail, campaign, course). One shared view for administrators and managers, with CSV export.
The content hash is what separates this from a log line: it proves not merely that a release happened but what was released. A later edit no longer matches the hash, and therefore shows.
Across all five products — LEGALinhouse, PUBLISHinhouse, RECEPTIONinhouse, VOICEinhouse and ACADEMYinhouse.
A record that shows only the reviewed cases is marketing.
Where a send happens automatically — a rule replying on its own, a scheduled run — it is flagged “without human”. Not hidden, not omitted, not a gap in the log, but an explicit label.
That is built in deliberately. An organisation that does not know its automated paths cannot answer for them — and a supervisory authority that finds them afterwards finds them in the worst possible situation.
Art. 50(4) sub-paragraph 2 requires deployers to disclose that a published text was artificially generated where it informs the public on matters of public interest.
That duty falls away where the content “has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication.” That is precisely what this record evidences: not just compliance hygiene, but the documented basis for an exemption you would otherwise have to assert unproven.
Art. 50(1) additionally requires people to be told they are interacting with an AI system unless that is obvious, and Art. 50(5) requires that information to be clear, unambiguous and accessible at first interaction.
Art. 14 (human oversight) applies only to high-risk AI systems — every paragraph of the article says so. We therefore do not claim this record “satisfies Art. 14”; that would imply your deployment is high-risk, which in most cases it is not.
The honest statement is stronger: if one of your use cases does fall into the high-risk band — which you determine yourself in the LEGALinhouse AI Act register — Art. 14(4) requires your overseers to understand the output, interpret it correctly, override or reverse it, and stay alert to automation bias, the tendency to over-trust an AI output. This record is the evidence that supports exactly that.
Accountability requires you to demonstrate compliance. A timestamped record with a person and a content hash demonstrates. A process description does not.
Most AI tools build the human in as an interaction step — a button. CEAVEO builds the human in as evidence.
The difference surfaces the moment someone asks: show me that this letter, this article, this customer reply was released by a person. A button leaves nothing behind. A hashed record answers in one export.
Together with pervasive pseudonymization, the AI Act register and the two-tier RDG model, that makes a compliance architecture built from artefacts rather than assurances.
Positioning: the record documents your review — it does not replace it and does not assess it. Responsibility for a release stays with the person granting it.
The platform is in closed beta; general availability November 2026. Access on request.