An audit right in AI contracts: specify the evidence

You must now specify in every AI contract which inspection evidence the supplier will provide, within what deadline, with what access powers, and what remedies apply if they refuse or the evidence is incomplete. A general audit clause does not guarantee you can verify a concrete AI outcome.

An analysis of 26 October 2026 of the specification of inspection evidence in AI contracts argues that a contractual audit right naming only general documentation does not ensure you can reconstruct a specific AI decision afterwards. The European Commission's Expert Group on AI Contracting, which began work in July 2026, is developing model contract terms for AI procurement to reduce practical risks. In our assessment, this makes the design of audit and verification rights a concrete contractual question now, not a theoretical preference: the Commission provides building blocks for purchasers to assemble, not a universal clause you can adopt without modification.

What does the AI Act actually require here?

Article 25 of the AI Act obliges the provider of a high-risk AI system and any third party supplying it with a model, tool, service, component or process to specify by written agreement the necessary information, capabilities, technical access and assistance needed to comply with the regulation. The guidance on Article 25 emphasises that this is an obligation to specify access, not an unlimited standalone right to inspect everything. The Commission's own guidance on the AI Act underlines that high-risk systems must be traceable and auditable and that appropriate documentation must be maintained. In our reading, this means your contract should specify the relevant records, versions and retention periods so you can test traceability in practice. A contract that does not enforce this makes traceability depend on the supplier's goodwill.

Why does a general compliance statement not work?

A peer-reviewed study examining the contractual vacuum in AI procurement describes how AI-as-a-Service can deprive a purchaser of the telemetry needed to explain and correct automated actions. Because several suppliers often contribute to one output, the audit right must run through to the underlying supplier that carried out the disputed step. Where an audit right stops at the direct supplier, the records held by its sub-processors may remain out of reach, even though that is precisely where the evidence can lie. In our assessment, this is a significant blind spot in audit arrangements: a contract that does not enforce transitive audit rights leaves you unable to reconstruct decisions that depend on multiple suppliers' contributions.

Which concrete controls must your contract specify?

1. Name the specific records the supplier must produce — logs, model versions, tool-usage records, training data provenance and decision traces relevant to your use case.
2. Set a retrieval deadline and retention period — specify how quickly the supplier must make records available and how long they must keep them.
3. Define your inspection powers — state whether you can sample, audit sub-processors, examine model versions, and inspect the supply chain for AI components.
4. Require evidence of completeness — the supplier must confirm that the records provided are complete for the period and scope you requested, and must record any gaps.
5. Establish remedies for non-compliance — include suspension, step-in rights and termination if the supplier refuses access, provides incomplete evidence or cannot reconstruct the outcome.

What failure modes does this protect against?

• Sub-processor opacity — audit rights that stop at the direct supplier leave you unable to reach the records held by their sub-processors.
• Incomplete telemetry — suppliers may provide compliance statements without the specific logs or model versions needed to reconstruct a decision.
• Trade secret claims — suppliers can fend off verification by invoking confidentiality unless the contract specifies remedies for refusal.
• Version drift — without retention periods, suppliers may delete or overwrite the model versions or training data that produced a specific output.
• Sampling gaps — a general right to audit does not guarantee you can verify the specific decisions or populations you need to explain.

How should you build this into your next contract?

An enforceable clause names not just an "audit right" but the records and access the supplier must actually produce. Build your clause along five elements: the specific records and artefacts relevant to your use case; the deadline within which the supplier must provide them; your inspection powers, including access to sub-processors and model versions; the scope and period you can audit; and the remedies—suspension, step-in, termination—that apply if evidence is missing or access is refused. Test the completeness of the evidence through sampling, period, version history and traceable records, and set out how missing evidence is recorded. That distinction between a contractual right to inspect and the concrete records that must be available determines whether you can turn a technical AI explanation into usable decision evidence.

Model terms, the AI Act and the academic framework must be aligned with the risk, the sector, the applicable legislation and the actual system architecture. No single universal audit clause will fit every contract. The lawyer must now include an inspection-evidence list in every new contract rather than an open audit sentence; the CISO must determine before signing which logs, model versions and tool-usage records are retrievable and with what retention period; the contract manager must negotiate a transitive audit right covering relevant sub-processors; and the director must have suspension, step-in and termination on non-reconstructability included as hard sanctions. Tooling can help you track which records you have requested, which the supplier has provided, and which gaps remain. Your own professional judgement must decide which records matter for your use case, which sub-processors are material to your risk, and which remedies are proportionate to the decisions at stake.