What the ICO published on 11 November
On 11 November 2026, the UK Information Commissioner's Office published guidance on subject access requests generated by artificial intelligence. The guidance covers a pattern regulators are now seeing often. A person asks a chatbot to write a request for their personal data and sends the chatbot's draft to the organisation. The ICO's position is simple: "You must treat an AI-assisted SAR in the same way as any other SAR."
The ICO guidance goes into practical detail. A request stays valid even when the requester's AI tool cites the wrong law or invents case law. An organisation can ask for clarification when a request is unclear or very large, and the response clock pauses while the organisation waits for an answer. The organisation can't force the requester to narrow the request, though. Using AI to draft a request doesn't on its own make the request manifestly unfounded or excessive. The ICO also warns that AI tools can produce convincing fake identity documents and fake letters of authority. Where a letter of authority looks generated, the ICO suggests contacting the person the request is about directly. Metadata that identifies the person falls within scope. An organisation doesn't automatically have to search every type of metadata, but it should do so where the metadata is likely to hold relevant information or the requester asks for it.
Why the UK guidance applies in New Zealand
The ICO guidance has no legal force here. NZ businesses still receive the same kind of AI-written requests under information privacy principle 6 of the Privacy Act 2020, and the NZ rules lead to much the same answers. On 29 September 2026, Bell Gully wrote that AI-drafted requests to NZ organisations tend to be long and legalistic. Those requests often ask for every category of personal information and try to head off withholding grounds in advance, and they can trigger quick follow-up questions. Bell Gully's view is that a request doesn't become vexatious because an AI wrote it, or because answering it is broad work or costly.
The Office of the Privacy Commissioner's own guidance supports that reading. Section 53(h) lets an organisation withhold information where the request is frivolous or vexatious or the information is trivial. The OPC says the ground is rarely used and that the request itself has to be vexatious. A difficult or even malicious requester doesn't meet the test. An OPC case note about an airline passenger shows how the test worked in practice. The passenger's first request was legitimate. Only his later requests qualified, because they came with abusive calls, ignored the airline's attempts to resolve the issue, and came with a stated plan to make the airline look as bad as possible.
Two other NZ obligations matter more once AI-written requests start arriving. The OPC says organisations have 20 working days to respond to a request. The same OPC page explains that section 57 requires an organisation to verify the identity of the requester, or of the person acting for the requester. Generated ID documents and authority letters make section 57 checks harder for a small privacy team that has relied on a quick look at a scanned driver licence.
Where the pressure lands for NZ businesses
For most owners, the legal tests are easier to handle than the workload. An AI-drafted request for "all personal information, including metadata" asks the business to search email, shared drives, CRM notes, ERP records, call recordings, ticketing systems and, more and more, the prompts and outputs that staff AI tools store. In its generative AI guidance, published in June 2023, the OPC said organisations using generative AI should develop procedures for responding to access and correction requests. The same OPC page warns that generative AI tools may not always be compatible with those rights. Many firms have since rolled out AI assistants without working out how anyone would search the assistants' chat histories within 20 working days.
The Privacy Act does give organisations some room. Bell Gully points out that an organisation only has to make reasonable efforts to find information, and doesn't have to supply information that isn't readily retrievable. Organisations can charge for supplying information, but the OPC treats charges as the exception. Those limits only help a business that can show where its personal information is stored and how hard a search would be. A business with a current map of its systems can quickly scope a broad request, explain the scope to the requester and write down its reasons. That's the approach the ICO describes for UK organisations.
NZ operations teams should take the ICO's fraud warning seriously. Suppose a request arrives with a fake authority letter and the business sends a customer's file to an impersonator. The business has then disclosed personal information to the wrong person, which can become a notifiable privacy breach. Access requests usually arrive first in a customer service inbox, and staff there rarely have any training in spotting a generated document.
The OPC hasn't published guidance aimed at AI-drafted requests, so for now NZ businesses are left to apply the existing principles and case notes themselves. The ICO note gives privacy teams a useful checklist in the meantime, because its conclusions line up with the OPC's published position on vexatious requests.
Next steps
- Log every access request when it arrives, whatever its format. The OPC airline case note records that the Privacy Act doesn't require written requests. The airline used up the first 20 working days waiting for a written request it wasn't entitled to demand.
- Set up a short triage routine. Identify the core of the request, contact the requester to confirm what they want, and record any scope you agree. Asking for clarification is fine, but treat any narrowing as the requester's choice.
- Map where personal information sits, including AI tool histories, meeting transcripts and audit logs, and note who can search each system. Our AI governance and adoption reviews should cover access requests for every AI tool a business uses, and so should yours.
- Decide in advance how the business will handle metadata and system logs when a request is broad, and write down the reasoning.
- Tighten identity and authority checks under section 57. When a letter of authority looks generated or doesn't match your records, contact the customer directly using the details you already hold.
- Treat access requests as a service process with an owner, a deadline and a queue. Firms that already run ITIL 4 request management can send privacy requests through the same tooling with a 20 working day target.
- Record the reasons for any refusal under section 53(h), and never give AI drafting as a reason.
Sources
- Subject access requests generated by artificial intelligence (ICO)
- ICO Guidance on AI-Generated Subject Access Requests Explained (Captain Compliance)
- How AI is reshaping privacy access requests (Bell Gully)
- Frivolous, vexatious or trivial (Office of the Privacy Commissioner)
- Passenger's vexatious requests to airline (Office of the Privacy Commissioner case note)
- Confirming a requester's identity (Office of the Privacy Commissioner)
- Generative artificial intelligence (Office of the Privacy Commissioner)
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help