A supplier to a central government department described the moment the requirement arrived. Not through a regulator, not through a consultation, and not with notice. It arrived as three questions inside a tender response template, with a submission deadline eleven days away, asking how the supplier governs the AI in its product, what documentation it can provide, and what human oversight the department would need to apply.
That is how public sector AI governance reaches suppliers. The obligations sit with the department. The department discharges them partly through the people it buys from, so the requirement travels down the supply chain as procurement documentation rather than as regulation. Suppliers meeting it for the first time in a tender have days rather than months to answer, and the answer is scored.
The question moving down the supply chain
The asymmetry is worth stating. A department has policy teams, a data ethics function in some cases, and time to develop a position. A supplier encountering the same expectation in a tender has a word limit and a deadline.
The consequence is predictable. Suppliers write something new under pressure, describing governance that is partly real and partly aspirational, and it enters the record. If the bid is won, the department may then ask for the evidence behind the answer, and the gap between what was written and what exists becomes a contract management conversation.
The suppliers that handle this well are not the ones with the best writers. They are the ones that already hold the inventory, the impact assessments, and the oversight design, and can assemble a response from material that already exists.
The Algorithmic Transparency Recording Standard
Central government departments and arm's length bodies have been required to publish records for algorithmic tools meeting the applicable criteria. A record covers what the tool does, why it is used, how it works in outline, what data it relies on, what oversight and review arrangements apply, and who is accountable.
The Algorithmic Transparency Recording Standard was mandated across all government departments on 6 February 2024, with the Mandatory Scope and Exemptions Policy published on 17 December 2024. It applies to all government departments and to arm's length bodies delivering public or frontline services or directly interacting with the general public, covering algorithmic tools that significantly influence public decision-making or directly interact with the public.
The supplier implication is direct and frequently overlooked. A department publishing a record about a tool you supplied needs the underlying information from you, in a form suitable for publication. That means descriptions of the model, the data, and the oversight arrangements that are accurate, plain enough for a general audience, and cleared for public release. A supplier that treats all of this as commercially confidential creates a problem the department has to solve, and departments notice which suppliers make their obligations harder.
Preparing a publication-ready description in advance is a small piece of work. Producing one under a departmental deadline, with a legal review of what can be disclosed, is not.
Where AI governance appears in public procurement
Three places: framework agreements and their associated questionnaires, tender-level selection and award questions, and contract conditions including schedules on data, ethics, and change control.
Public procurement questions tend to be more prescriptive than commercial ones and considerably less negotiable. A commercial buyer will often accept a mapped answer, a policy extract, and a call with your security lead. A public buyer asks a specific question, applies a published scoring methodology, and cannot informally trade a weak answer against a strong relationship.
Government guidance on the use of AI in the public sector has continued to develop, including material on the procurement of AI and on responsible use, and the direction of travel has been towards more explicit expectations of suppliers. The status of individual documents varies between guidance and requirement, so suppliers should check what applies to a specific route to market rather than assuming a uniform standard.
The accountability question
Public bodies carry duties that commercial buyers do not. The public sector equality duty requires due regard to the effects of decisions on people with protected characteristics. Decisions are subject to judicial review, where the question is whether the decision was lawfully and rationally made, and whether the process was fair.
An automated or assisted decision affecting an individual creates exposure the department cannot delegate to a vendor. If a decision is challenged, the department has to explain how it was reached. "The supplier's system produced it" is the beginning of a difficult exchange, not the end of one.
The practical consequence for suppliers follows from that. Departments increasingly require documented evidence of how the system works, what it was trained on, what its limitations are, where human judgement is exercised and with what authority, and how performance is monitored after go-live. They require it because they need it for their own defensibility, which means the requirement does not soften with a good relationship.
Where the AI sits in public sector delivery
Case triage and prioritisation across casework functions. Eligibility assessment and benefits processing support. Fraud and error detection. Document processing, correspondence handling, and summarisation. Resource allocation, scheduling, and demand forecasting. Citizen-facing chat and service navigation.
Several of these map onto EU AI Act high-risk categories where the supplier also operates in the EU, including access to essential public services and benefits, law enforcement, and migration. Those Annex III standalone obligations apply from 2 December 2027, following the Digital Omnibus amendment given final approval by the Council on 29 June 2026. A supplier selling the same product on both sides of the Channel should classify per market rather than assuming the UK position governs.
What ISO 42001 gives a public sector supplier
A documented AI system inventory in a form that can feed transparency reporting rather than needing translation first. Impact assessment covering affected citizens, including those least able to challenge an outcome. Human oversight design with documented authority, specifying what the department must do and what the product enforces. Data governance and lineage, so the training data question has an answer. Monitoring and performance evidence over time, which is what turns an assertion about accuracy into something a department can rely on. And a management system that answers procurement questions from an existing evidence base rather than a scramble against a deadline.
The limit needs stating as clearly here as anywhere. ISO 42001 certification is not regulatory compliance. It is not a public sector requirement, it does not discharge a department's obligations, and no framework grants it standing. Its value is that it produces the evidence the questions ask for, checked independently, in a structure a buyer recognises.
Where clearance and defence supply chain requirements intersect
For suppliers operating in defence or handling OFFICIAL and OFFICIAL-SENSITIVE material, AI governance arrives alongside a set of cyber requirements that are already familiar: Cyber Essentials and Cyber Essentials Plus, ISO 27001, JOSCAR registration, and contract-specific security conditions.
The practical point is that these now appear in the same questionnaire. A supplier with mature cyber evidence and nothing on AI governance answers half a question, and the half that is missing is the half the buyer is least able to waive, because it is newer and there is no precedent for accepting a gap.
Personnel security matters in the same conversation. Where delivery requires cleared personnel, the governance question extends to who can access the models and the data used to develop them, and that has to be answerable in the same document set.
Sequencing for a public sector supplier
Inventory and classify. Establish, per tool, what it does, who it affects, and where a decision about an individual is being made or influenced.
Establish whether transparency reporting applies to any tool you supply, and what the department will need from you if it does. Ask the department rather than guessing, and ask before the tender rather than after.
Build the management system, using the existing information security spine where one already operates, since most public sector suppliers hold ISO 27001 or an equivalent.
Produce the department-facing documentation as a specific deliverable with an owner and a review date, not as an internal artefact reissued on request.
Certification where the procurement case supports it, which in this market usually means where a framework renewal or a specific pursuit puts a value on it.
An AI governance readiness assessment establishes the inventory, the classification, and the documentation gap against procurement and transparency expectations, so the next tender is assembled rather than written. Book a scoping call.