The disclosure lands on the tender itself. Annex B of PPN 017 carries example wording a buyer can drop into procurement documentation, asking suppliers to disclose their use of AI in the creation of their tender [9]. Where AI sits inside the product being sold is handled elsewhere: the note tells commercial teams that a range of guidance exists on AI as a subject area, on appropriate use of AI in public services and on how AI products should be procured, with references in Annex A [15].
A bid function reading that question will treat it as a box with a yes and a sentence of reassurance attached. The buyer's instructions run the other way. Where suppliers use AI tools to create tender responses, the note says additional due diligence may be required to establish that the supplier has the capacity and capability to fulfil the contract, proportionate to the specific risk, and it names site visits, clarification questions and supplier presentations as the available forms [11]. The stated purpose is to test the accuracy, robustness and credibility of the tender, the way a contracting authority would chase any other ambiguity in a bid [12].
The clause with teeth is the one about training data. Buyers are told to put proportionate controls in place so suppliers do not use confidential contracting authority information, or information not already in the public domain, as training data, and the worked example is confidential government tender documents fed to an LLM to produce future tender responses [10].
Buyers pick up a workload instruction as well: plan for a general increase in activity as suppliers use AI to streamline bid writing, which shows up as more clarification questions and more tender responses [13], and consider allowing more time in the procurement to absorb the volume and the due diligence [14].
The gap between the note's original issue and the date its provisions took effect is fifteen months [17].
Before anyone agrees disclosure wording for a public-sector bid, two questions decide how that wording should be handled. Whether AI touched the words in the submission, and who can say so with confidence about the whole document rather than their own section, sets how much delivery proof the buyer may ask to see face to face. Whether anything from a contracting authority that was not already public has gone into a model the company trains, tunes or retrieves from is the costlier question, because disclosure wording can still be redrafted before the deadline, while whatever a model has already absorbed stays absorbed no matter when the deadline falls.
