Auditrek vs OneTrust AI Governance: what we can verify
If you are comparing Auditrek with OneTrust AI Governance, start with the rule this page follows: we describe OneTrust only in its own published wording from its product page, retrieved 16 November 2026, and everything we cannot source we leave as a question for you to put to them.
Auditrek is not open yet, so anything on this page about what it does is what it is being built to do.
Where each product sits
Both products are bought by someone who will eventually have to answer a single question from an auditor or a supervisory authority: show me how this system reached that outcome, and show me the record has not been altered since it was written.
OneTrust AI Governance states its own position as "Ship AI Faster With Risk Control Where AI Runs", and describes its scope as "continuous discovery, inventory, monitoring, and policy evaluation across models, data, agents, and vendors" and as the ability to "Detect issues in real time, apply guardrails consistently, and generate audit-ready evidence with runtime enforcement, attestations, and ongoing validation". Those are quotations from its product page, retrieved 16 November 2026, and we quote rather than summarise because product taxonomies in this category change often and because it is not our place to characterise someone else's scope.
Auditrek starts one layer down, at the individual inference. The REST and webhook API accepts a decision record per event, carrying the input hash, the model id and version, the output, the confidence, the timestamp and the human override flag, and stores those records as a SHA-256 hash-chained append-only sequence per registered AI system. Each record's hash is computed from the previous hash and the record payload, so a sequence can be recomputed end to end rather than trusted on the strength of a database row. From that sequence, the report generator produces PDF and HTML output for a chosen system and a chosen date range, with fields mapped to EU AI Act Article 12 and Article 13.
The article numbers matter here because they describe different obligations. Article 12 requires that "High-risk AI systems shall technically allow for the automatic recording of events (logs) over the lifetime of the system". Article 13 requires that such systems "shall be designed and developed in such a way as to ensure that their operation is sufficiently transparent to enable deployers to interpret a system's output and use it appropriately". Auditrek is built to support the documentation of those obligations by recording what happened. It does not assess whether your practices satisfy them, and no report it generates asserts that they do.
What we quote from OneTrust, and what we leave open
Our house rule is that a claim cites a mechanism, an article number or a measured outcome. Applied to a competitor, that means we hold ourselves to published wording and nothing else.
What we hold on OneTrust AI Governance is its product page, retrieved 16 November 2026. From it we can quote the positioning above, the description of "continuous discovery, inventory, monitoring, and policy evaluation across models, data, agents, and vendors", the statement that the platform can "Turn AI policy into controls that work across homegrown and third-party AI. Detect shadow AI, prevent sensitive data leaks, and give security and governance teams proof that controls are working in production", and the promise to "generate audit-ready evidence with runtime enforcement, attestations, and ongoing validation". Its published capability list also names built-in templates for EU AI Act, NIST AI RMF and ISO 42001 assessments, and native connections with Amazon Bedrock, Microsoft AI Foundry, Google Vertex and Databricks, plus Jira and Palo Alto Networks. That same page publishes no pricing and directs visitors to sales. Independently of the vendor, one write-up of Gartner's inaugural Magic Quadrant for AI Governance Platforms records that "Visionaries included Airia, Credo AI, ModelOp, Monitaur, and OneTrust", retrieved 16 November 2026.
What we do not hold is anything on the internal structure of their records, their verification method, their identity options, their hosting regions or their prices. So the third column of the table below carries the quotation where one exists and the question where it does not. Fill it in from their mouth, in writing, not from ours.
If you are running a formal evaluation, ask both vendors for the same three artefacts and compare those rather than feature lists: a sample of the generated output for a real date range, the ingestion contract including the exact payload fields, and the exact wording and position of any disclaimer on documents you might later hand to an authority. Those three answers separate vendors faster than a grid does.
Side by side, with the unsourced cells left as questions
| Dimension | Auditrek | OneTrust AI Governance, in its own published wording (retrie |
|---|---|---|
| Smallest unit of record | One append-only decision record per inference event: input hash, model id and version, output, confidence, timestamp and human override flag, held per registered AI system. | Its page describes governing "across models, data, agents, and vendors" and giving teams "proof that controls are working in production". We hold no wording on the smallest stored object, so ask wheth |
| Integrity mechanism | SHA-256 hash chaining, each record's hash computed from the previous hash and the record payload, written in one transaction with its sequence number. | Its page describes "audit-ready evidence with runtime enforcement, attestations, and ongoing validation". The wording we hold does not describe cryptographic chaining either way, so ask what makes a s |
| How a record is verified | A server-side endpoint walks the chain for that system from sequence one, recomputes each hash from the stored previous hash and payload, and returns a tamper status with the first broken sequence num | Not established from any wording we hold. Ask whether verification recomputes from stored data at the time you ask, or reports a status recorded earlier. |
| How events arrive | Generic REST and webhook ingestion, with documented payload mapping guides for AWS SageMaker and Azure ML and copyable examples in the API docs. No bespoke platform SDKs at launch, and nothing to inst | Its published capability list names native connections with Amazon Bedrock, Microsoft AI Foundry, Google Vertex and Databricks, plus Jira and Palo Alto Networks, under the positioning "Ship AI Faster |
| Sensitive fields at ingestion | A data-minimisation gate inspects each payload server-side and rejects recognisable raw sensitive fields with a 422 before any database write, and subject identifiers are pseudonymised by default. | Its page states the product can "Detect shadow AI, prevent sensitive data leaks". Beyond that we hold nothing, so put a concrete payload to them and ask what the system does with each field. |
| Precondition for connecting a system | No API key and no webhook secret is issued for an AI system until a DPA confirmation with a signed_at value is recorded, checked server-side and refused with a 403 if unsigned. | Not established from wording we hold. Ask whether a signed data processing agreement is a technical precondition to connecting a pipeline or only a contractual one. |
| Generated output | Audit report generator producing PDF and HTML for a chosen system and date range, with sections following EU AI Act Article 12 and 13 field names. One regulation and one format set at launch. | Its page offers to "generate audit-ready evidence", and its capability list names built-in templates for EU AI Act, NIST AI RMF and ISO 42001 assessments. Ask for a sample, then check which fields map |
| How the document is framed | Every page of every generated report carries the fixed footer: this report does not constitute legal advice and has not been reviewed by qualified regulatory counsel. | No wording available to us. Request the exact disclaimer text that appears on documents the tool generates, and where on the document it sits. |
| Sign-in | Email and password, with multi-user access inside an organisation. No single sign-on. | Outside what we can verify. Confirm which identity and access options are included at the tier you are actually quoted. |
| Price and billing basis | Published: Starter at €1,500 a month for two connected systems and 50,000 decision records a month, Growth at €3,500 a month for five systems and 500,000 records. Both declare a 30-day free trial with | Its product page publishes no pricing and directs visitors to sales. We will not estimate a figure. Treat it as a sales conversation until you have a written quote. |
| Hosting and residency | Amazon Web Services, in a single region. No multi-region replication and no choice of data residency, stated plainly in the app itself. | Not verified by us. Ask where data is stored and processed, and what residency options exist for the obligations you are working to. |
What Auditrek is built to do, in specifics
Capture the decision, not a summary of it
The REST and webhook API accepts inference event payloads with input hash, model id and version, output, confidence, timestamp and human override flag, stored as an append-only record per AI system.
Chain each sequence with SHA-256
Each record's hash is computed from the previous hash and the record payload and written with its sequence number in one transaction, so the sequence for a system can be recomputed end to end.
Verify by recomputation, not by a stored flag
The verification endpoint walks the chain from sequence one, recomputes each hash, and returns a tamper status with the first broken sequence number when the recomputation disagrees with what is stored.
Gate sensitive fields at the door
Ingestion rejects payloads carrying recognisable raw sensitive fields before any write, and pseudonymises subject identifiers by default, so raw model inputs never arrive in the first place.
No credentials before a DPA is recorded
A system cannot be connected and no webhook secret or API key is issued until a DPA confirmation with a signer name, document reference and signed_at is recorded, enforced on the server.
Reports in PDF and HTML, with the disclaimer fixed
Pick a system and a date range and the generator produces output following Article 12 and 13 field names, with the non-legal-advice footer embedded on every page and past reports kept exactly as generated.
One dashboard for the officer who has to answer
Decisions logged, systems connected, last report generated and current tamper-verification status, written for a compliance reader rather than for the engineer who wired up the ingestion.
When to choose which
This is not a grid you win on row count, because the two products are answers to different gaps.
Choose a broad governance platform when the gap is the programme
If what you are missing is the register itself, the classification of which systems fall in scope, the intake and approval workflow, the policy library, the vendor and shadow AI discovery, then the thing to buy is a platform that does that work, and OneTrust AI Governance is sold into exactly that need in its own published terms. Auditrek has none of it. We do not maintain a control library, we do not score risk, we do not route approvals, and we are not a place to store your assessments. A shortlist that treats a programme platform and a per-decision ledger as substitutes will disappoint someone later in the year.
Choose Auditrek when the gap is the per-decision record
The register tells an examiner which systems exist and what you concluded about them. A hash-chained ledger tells them what system seven output at 14:03 on 12 March, what the confidence was, whether a human overrode it, and whether the stored sequence still recomputes to the same hashes. If the request you are dreading is the second kind, that is the whole of what Auditrek is built for, and the reason it fits alongside rather than instead of a wider platform is that it takes events over REST or a webhook after the fact and needs nothing installed in your production inference path.
If you are considering both
There is no technical reason the two cannot coexist, and we make no claim about any integration between them, because none exists. Auditrek is not yet open to general sign-up, so the test below is one to plan rather than to run today: put one real pipeline through the ingestion with a payload you have actually mapped, generate a report for a date range an auditor might pick, and read what comes out. Then ask the same of any platform you are also considering, and compare the two documents rather than the two sales decks. Both plans declare a 30-day free trial with no card at trial start, which is the window that evaluation is meant to fit into once access opens.
If you want the longer version of that test, the questions to put to any vendor in this category are written up separately, and the sibling comparisons cover the other names a mid-market shortlist usually carries.
Questions
Does Auditrek replace OneTrust AI Governance?
No. Auditrek covers decision capture, hash chaining, server-side verification, a compliance dashboard and the Article 12 and 13 report, and nothing else. If your gap is inventory, intake, risk scoring, policy controls or shadow AI discovery, that is the category OneTrust AI Governance sells into on its own product page, and Auditrek does not compete for that work.
Can we run Auditrek alongside a platform we already pay for?
Yes, in the sense that Auditrek takes inference events over its own REST endpoint or a signed webhook and needs nothing installed in your production inference path. We claim no integration with any governance platform, including OneTrust, and would not describe one that does not exist.
What does Auditrek cost compared with OneTrust AI Governance?
Auditrek publishes its prices: Starter at €1,500 a month for two connected systems and 50,000 decision records a month, Growth at €3,500 a month for five systems and 500,000 records, each declaring a 30-day free trial with no card at trial start. Sign-up is not open yet, so those are the terms you will meet when it is. Payment runs through Stripe, which works out VAT at checkout. The OneTrust AI Governance product page we retrieved on 16 November 2026 publishes no pricing and directs visitors to sales, so ask them for a written quote rather than taking a figure from a comparison site.
Does an Auditrek report prove we meet EU AI Act obligations?
No, and we will not say otherwise. The report records what a chosen system did over a chosen date range, with sections following Article 12 and 13 field names, and every page carries a fixed footer stating that it does not constitute legal advice and has not been reviewed by qualified regulatory counsel. Whether the evidence is sufficient for a given examination is a judgement for your counsel and the authority, not for a tool.
How do we check a record has not been altered?
Call the verification endpoint for a record. It walks that system's chain from sequence one, recomputes each hash from the stored previous hash and payload, and returns a tamper status, naming the first broken sequence number when the recomputation disagrees with what is stored. The dashboard shows the latest verification status per connected system.
What happens if a payload contains a raw sensitive field?
Ingestion inspects the payload server-side for recognisable raw sensitive-field patterns and rejects the request with a 422 and an error body before anything is written, and subject identifiers are pseudonymised by default. Raw model inputs are not sent to Auditrek at all, only an input hash.
Where is the data held, and how do we sign in?
On Amazon Web Services, in a single region, with no multi-region replication and no choice of data residency. Sign-in is email and password, with multi-user access inside an organisation. There is no single sign-on, and there is no independent security certification such as SOC 2 or ISO 27001 in place today.
Who is behind Auditrek?
The product is Auditrek. Its terms and privacy pages, which are where the company that sells it is named with its registered details, are not published yet while the product is in its waitlist stage, and we would rather say that plainly than point you at a page that does not exist.