Auditrek vs Trustible: how to run the comparison honestly
If you are weighing Auditrek against Trustible, the short version is that the two sit at different levels of the same problem: Trustible's own homepage positions it as "The AI governance platform built for the intelligent enterprise" with a "Full audit trail generated as work happens", while Auditrek records one SHA-256 hash-chained decision record per inference event and generates PDF and HTML…
Auditrek is not open yet, so anything on this page about what it does is what it is being built to do.
Two questions, one budget line
Two different questions usually get funded from the same budget line. The first is the programme: which AI use cases exist, which obligations attach to each one, who owns them, which assessments have been done and which framework each control satisfies. The second is evidence at the level of a single model decision: what a given model produced on a given date, what its confidence was, whether a person overrode it, and how you would show that the record has not been edited since it was written. Both are real work. They are not the same purchase.
Trustible describes itself on its own homepage under the line "AI CLARITY IS VELOCITY" as "The AI governance platform built for the intelligent enterprise", and says it "gives enterprises the clarity, velocity, and confidence to govern AI across its entire lifecycle". The same page names "Structured intake routes every use case by risk automatically", "Map controls once and satisfy every framework", "Full audit trail generated as work happens" and "Audit-ready evidence always on, never a project", names support for "EU AI Act, NIST AI RMF, ISO 42001" plus "US, global, and industry-specific frameworks", and states "10x faster AI intake" and "4x more use cases approved". Those are their words, retrieved on 16 November 2026 and listed in the sources below. Their site is the only authority on what they currently ship, and everything else on this page about them is a question we would ask in a demo rather than a claim we are making.
Auditrek is deliberately narrower. It captures what your models actually did, one record per inference event. The API accepts an input hash, a model id and version, the output, a confidence value, a timestamp and a human override flag, over REST or webhook, and stores each event as a SHA-256 hash-chained, append-only record scoped to a single ai_system. From those records it generates a PDF or HTML report for one system and one date range, mapped to EU AI Act Article 12 and 13 fields, with a non-legal-advice disclaimer embedded in every report. One regulation and one format set at launch.
The collision is the phrase, not the product
Both products use the words "audit trail", and that is where shortlists go wrong. The question that separates them is what one row is. In Auditrek, a row is a single inference event, and the chain is recomputed on demand to say whether it has been altered. Read Trustible's own wording again: the trail is "generated as work happens", which describes the governance work being recorded. If you are deciding between these two, put that question first in both conversations and ask to see one row of each, with the field list, on screen.
Dimension by dimension
| Dimension | Auditrek | Trustible, in its own published wording |
|---|---|---|
| Primary unit of record | One decision record per inference event, submitted over REST or webhook with input hash, model id and version, output, confidence, timestamp and human override flag, scoped to one registered ai_system | Homepage wording is about governance work: "Structured intake routes every use case by risk automatically" and "Full audit trail generated as work happens" (trustible.ai, retrieved 16 November 2026). |
| How integrity is established | Records are SHA-256 hash chained and append-only. On each write the service reads the previous record's hash for that system, hashes the previous hash with the payload, and stores the fields in one tr | We retrieved no wording on their homepage describing cryptographic record integrity; the published claim is "Audit-ready evidence always on, never a project" (retrieved 16 November 2026). Ask how reco |
| Framework breadth | One regulation at launch: report fields mapped to EU AI Act Articles 12 and 13. No NIST or ISO mapping in current scope. | Homepage names "Map controls once and satisfy every framework" and names support for "EU AI Act, NIST AI RMF, ISO 42001" plus "US, global, and industry-specific frameworks" (retrieved 16 November 2026 |
| Report output | PDF and HTML for a chosen system and date range, with Article 12 and 13 field headings and a summary of the decisions in range. Every report carries a visible footer stating that it does not constitut | Not stated in the wording we retrieved beyond "Audit-ready evidence always on, never a project". Ask for a generated sample document and read its footer as well as its body. |
| Getting data in | Generic REST and webhook ingestion with documented payload mapping guides for AWS SageMaker and Azure ML, published on an API docs page with field definitions and worked examples. No bespoke platform | Not stated in the wording we retrieved. Ask which systems they read from, and whether evidence arrives by API, by connector or by a person filling in a workflow step. |
| Data handling at ingestion | A data-minimisation gate inspects each payload for recognisable raw sensitive field patterns and rejects the request before any database write. Subject identifiers are pseudonymised by default, and ra | Not stated in the wording we retrieved. Ask what their ingestion layer accepts, what it refuses, and where that is enforced. |
| Gate before a pipeline connects | No API key and no webhook secret is issued for a system until a DPA confirmation with a signed_at timestamp is recorded, checked server side on the connect endpoint. | Not stated in the wording we retrieved. Ask about their processing terms and at which point in onboarding they take effect. |
| Day-to-day screen | A web dashboard showing decisions logged, systems connected, last report generated and tamper-verification status, plus a paginated decision table with a per-record verify action, written for a non-te | Not stated in the wording we retrieved; the published outcome figures are "10x faster AI intake" and "4x more use cases approved" (retrieved 16 November 2026). Ask to be shown the screen your team wou |
| Sign-in and access | Auth-gated multi-user organisation access. People sign in with email and password. There is no single sign-on, and no SOC 2 or ISO 27001 certification and no independent cryptographic review of the ha | Not stated in the wording we retrieved. Ask about sign-in methods and assurance reports if your security review requires a specific one. |
| Price, published | Starter EUR 1,500 a month for 2 connected AI systems and 50,000 decision records a month. Growth EUR 3,500 a month for 5 systems and 500,000 records a month. Both carry a 30-day free trial with no car | No pricing is published on the landing page we retrieved, and the research we hold describes the motion as sales-led (retrieved 16 November 2026). Ask their team for current figures and for the unit o |
| Hosting and residency | Amazon Web Services, in a single region. There is no multi-region replication and no choice of data residency, and the dashboard carries a permanent notice saying so. | Not stated in the wording we retrieved. Ask where data is hosted and whether residency can be chosen. |
| Availability today | Waitlist. Auditrek is not generally available, and everything in this column is what the product is built to do and what its code enforces, not a record of deployments. | Their own site is the authority on availability and terms, and we will not characterise it beyond the published wording quoted on this page. Ask them directly. |
What Auditrek is actually doing under the hood
We describe mechanisms rather than outcomes because the mechanism is the whole argument. Verification is not a lookup against a flag someone could set. The endpoint walks the chain for that system from sequence 1 to the record you asked about, recomputes each hash from the stored previous hash and payload, and returns either a valid status or the first sequence number where the chain breaks. That is the difference between a row labelled immutable in a database and a record whose integrity can be recomputed in front of a sceptical reader.
Two consequences follow, and both matter when you are comparing tools.
- Nothing sits in the production inference path. Events are posted to Auditrek out of band, over REST or webhook, after your model has already returned its answer. There is no model migration, no proxy in front of the endpoint, and no new runtime dependency between your inference service and a compliance vendor. Your ML engineers will raise this in the first ten minutes of any evaluation, so it is worth knowing where each candidate stands before the meeting.
- The ledger is narrow on purpose. Auditrek holds decision records and generates documentation from them. It does not hold your use-case intake, your risk assessments, your conformity work, your obligation map or your policy library. If those are the gap, a programme platform is the right purchase, and Trustible's published wording is aimed squarely at that layer.
What we will not claim
The reports are descriptive by design. They record what a system did, when, and whether a human overrode it, in a structure that follows the Article 12 and 13 field sets, and they are built to support your documentation of those obligations. They do not assert that your practices satisfy any obligation, and the embedded disclaimer says as much on the face of every page. Whether a given evidence pack satisfies a supervisory authority is a judgement for your counsel.
For the same reason, here is the honest list of what Auditrek does not have: no single sign-on, no SOC 2 or ISO 27001 certification, no independent cryptographic review of the hash chain, no multi-region hosting or residency choice, and no standard data processing agreement offered by Auditrek for its own customers to sign. If any of those is a hard gate in your vendor review, you will find out faster reading it here than three weeks into a security questionnaire.
When to choose which
Lean towards a programme platform
If your open question is scope and process, buy the tool that answers scope and process. When nobody can currently list every AI system in the business, when the board paper needs a register with owners and assessment status, when you are being measured against more than one framework at once, or when intake is the bottleneck because every new use case queues behind a committee, then a platform built around intake, control mapping and assessments is what moves you forward. Trustible's own published emphasis on intake routing, control mapping across frameworks and evidence generated as governance work happens describes that job. A per-inference ledger is premature when you do not yet know which systems should be logging at all.
Lean towards Auditrek
If your open question is evidence at the decision level, the shape of the problem is different. You already know which systems matter. Someone, an internal audit team, a supervisory authority, a large customer's vendor review, has asked or will ask what a specific model decided on a specific date, what the confidence was, whether a person overrode it, and how you know the record has not been touched since. Workflow records and screenshots answer the governance question and leave the decision question open. A hash-chained record with a server-side recomputation is a cleaner answer to a hostile one. The industry language for this shift, from policy-based governance to evidence-based governance judged on verifiable records, is exactly the move being described. Our use-case page for the compliance officer who has to produce the evidence walks through that moment in detail, and the guide on choosing AI audit trail software gives you the eleven questions we would put to any vendor, including us.
Quite possibly both, or neither yet
These are not mutually exclusive line items, and an honest comparison should say so. Plenty of organisations will run intake and assessments in one system and decision-level records in another, feeding reports from the second into the files kept by the first. A good number will decide this year's budget covers one and not the other. The third option nobody sells is waiting, on the view that guidance and deadlines are still moving. That is defensible for some systems and expensive for others, and it is a judgement about your own exposure rather than a software choice. Our alternatives page lays out the wider field, including the build-it-in-house route on an immutable log.
How to run the comparison yourself
- Write the question you expect to be asked
Pick a real one. "Show me every automated decline this model produced in Q2, and show that the log has not been altered since." Take that sentence into both demos and watch which parts each tool answers and which parts land back on your team.
- Ask for a generated document, not a slide
Request an output from real or seeded data, for one system over one date range, in the format you would hand over. Read the footer as carefully as the body: what a tool says about the limits of its own output tells you how it will behave under scrutiny.
- Ask how integrity is demonstrated, not just claimed
Ask what happens if a row is edited directly in storage, and what the product returns afterwards. A recomputed chain and a stored flag look identical on a feature grid and behave very differently in an examination.
- Check what touches the inference path
Bring the engineer who owns the endpoint. Establish whether adoption means a proxy, an SDK, a model migration or an out-of-band POST, because that answer decides your timeline more than any feature does.
- Settle data handling before the pilot, not during it
If your pipelines carry credit, hiring or clinical data, agree what is allowed into the record and what is refused at ingestion, and get processing terms signed before any credential is issued. Auditrek enforces that order in code: no recorded DPA confirmation, no API key or webhook secret.
- Compare the published numbers, then the quoted ones
Auditrek publishes EUR 1,500 and EUR 3,500 a month with system and record limits attached. Where a vendor quotes rather than publishes, ask for the unit of billing and the limits in writing so you can compare on the same axis.
Questions
Is Auditrek a replacement for an AI governance platform?
No. Auditrek captures per-inference decision records and generates documentation from them, mapped to EU AI Act Article 12 and 13 fields. It does not maintain use-case intake, an obligation map, risk assessments or a policy library, and it covers one regulation at launch. If those are your gap, a programme platform such as the one Trustible describes is the right purchase, and the two can sit side by side.
What does Trustible say about its own audit trail?
Its homepage states "Full audit trail generated as work happens" and "Audit-ready evidence always on, never a project", alongside "Map controls once and satisfy every framework" and named support for "EU AI Act, NIST AI RMF, ISO 42001" plus "US, global, and industry-specific frameworks", retrieved 16 November 2026. That is their wording and their site is the authority on it. The question we would ask in the demo is what one row of that trail contains, and whether an individual model inference is a record.
Can Auditrek reports be presented during a supervisory examination?
They are structured to document what your systems did, following the Article 12 and 13 field sets, and every report carries a visible statement that it does not constitute legal advice and has not been reviewed by qualified regulatory counsel. Whether a particular evidence pack satisfies an authority is a judgement for your counsel, not for us.
Do we have to change our models or move our inference workloads?
No. Events are posted to Auditrek out of band over REST or webhook after your model has returned its answer, with documented payload mapping guides for AWS SageMaker and Azure ML on the API docs page. There are no bespoke platform SDKs in current scope and nothing sits in the production inference path.
What stops sensitive data ending up in the record?
A data-minimisation gate at ingestion inspects each payload for recognisable raw sensitive field patterns and rejects the request before any database write, and subject identifiers are pseudonymised by default. Raw model inputs never arrive, only an input hash. Separately, no API key or webhook secret is issued for a system until a DPA confirmation with a signed_at timestamp is recorded.
What does Auditrek cost, and is there a trial?
Starter is EUR 1,500 a month for 2 connected AI systems and 50,000 decision records a month. Growth is EUR 3,500 a month for 5 systems and 500,000 records a month. Both include a 30-day free trial with no card to start. Billing is per organisation through Stripe, which works out VAT at checkout, and card details stay with Stripe.
Does Auditrek support single sign-on, and does it hold a SOC 2 report?
No to both. People sign in with email and password, and organisation access is multi-user and auth-gated so compliance, risk and engineering can work in one workspace, which the sibling page on multi-user org access covers. There is no independent security audit or certification such as SOC 2 or ISO 27001, no independent cryptographic review of the hash chain, and hosting is on Amazon Web Services in a single region with no residency choice.