TEARDOWN Published 14 November 2026 at 08:10. Evidence-based. Source-cited. No sponsored content.

The ICO's new guidance on AI-written data requests lets an abusive tone count toward refusing one. Its own detailed guidance, still in force, says tone alone should never be enough.

4 out of 5 stars4/52 documented mistakes in this teardownHow ratings work

Estimated reading time: 6 minutes

Rows of archive shelving holding boxed paper records, St Helena Archives, Jamestown.
Archive shelving holding personal records, St Helena Archives, Jamestown, 12 March 2018. Photo: Luke McKernan / Wikimedia Commons, CC BY-SA 2.0.

In short. The ICO's new guidance on AI-assisted subject access requests (SARs) says a request that is "clearly malicious, abusive or designed to cause disruption, this may support a decision that it's manifestly unfounded or excessive" [1]. The ICO's own detailed guidance on exactly that test, last updated 8 December 2025 and still the standing authority the new guidance links to, says "aggressive or abusive language is not acceptable" but "the use of such language does not automatically make a request manifestly unfounded," and requires concrete markers, a stated intent to disrupt, a grudge against a named employee, a systematic campaign, before the threshold is met [2]. The new guidance also names a resourcing burden AI-assisted requests create, then offers organisations no funded support for it, only a suggestion that they build their own materials. Two documented mistakes: four stars.

A subject access request is a person's statutory right to see what personal information an organisation holds on them, and to understand how it is used. People are increasingly asking AI tools to help them write one, and on 11 November 2026 the ICO published its first dedicated guidance on how organisations should handle the result, "AI-assisted SARs" in the regulator's own term [1]. It is a short, practical document, structured as a long list of questions and answers, and it carries no review date and no enforcement mechanism of its own: this is advisory "should" and "must" language, not a statutory instrument, and it attaches to no consultation and no sunset clause.

Most of it is protective of requesters. "A request isn't invalid just because a person used AI to help draft it," the guidance states, and organisations "must treat an AI-assisted SAR in the same way as any other SAR" [1]. It repeats the point for complexity: "the use of AI doesn't automatically make a SAR complex," and a request is not complex "simply because it's lengthy, unclear, convoluted or includes legal references" [1]. On that reading, the guidance sets a deliberately high bar against organisations using "an AI wrote this" as an excuse to delay or narrow a response.

Where the guidance departs from its own authority

The exception is the one answer that decides whether a request gets refused altogether. Under "Tone, volume and repeated requests," the guidance states: "People may prompt AI tools to generate SARs that are abusive, malicious or deliberately difficult to interpret. You should assess the content and circumstances of the request in the usual way, even if they've used AI. If the SAR is clearly malicious, abusive or designed to cause disruption, this may support a decision that it's manifestly unfounded or excessive" [1]. That sentence links, as its only cited authority, to the ICO's detailed right of access guidance on when a SAR counts as manifestly unfounded or excessive [1].

That detailed guidance, last updated 8 December 2025, draws a careful line the new guidance does not repeat. "While aggressive or abusive language is not acceptable, the use of such language does not automatically make a request manifestly unfounded. You should consider all the circumstances of the request and take a reasonable and proportionate approach in deciding whether or not this is a relevant factor" [2]. It defines "manifestly unfounded" with specific markers: the person explicitly states an intent to cause disruption, makes unsubstantiated accusations prompted by malice, targets a particular employee over a personal grudge, or systematically sends requests as part of a campaign [2]. "This is not a simple tick-list exercise," the detailed guidance adds, and a request that otherwise "appears to be a reasonable and genuine attempt to exercise the right of access" is unlikely to qualify, however it reads [2].

The new guidance's one-line test, "clearly malicious, abusive or designed to cause disruption," drops the qualifier that does the actual work in the document it cites. It gives organisations facing an AI-drafted request that reads as formal, assertive or "abusive" in tone a shorter route to refusal than its own detailed guidance allows for every other SAR, at the exact moment the rest of the same document insists AI-assisted requests should be treated no differently.

The burden the guidance names and leaves unfunded

The guidance is candid, early on, that AI-assisted SARs create real costs. Such requests "can make SARs complicated and time consuming for you to respond to," and "may also place additional pressures on your resources" [1]. It adds a second, more specific burden for fraud risk: "If you think they may have generated a letter of authority using AI, you should contact the person the request is about directly" [1], a new verification step with no definition anywhere in the document of how an organisation is meant to recognise an AI-generated letter of authority, and no referenced tool or training for that specific judgement.

Having named both pressures, the guidance's only offered response is for the organisation to build its own materials: "train staff," "provide clear guidance on how to make a SAR," "publish optional example prompts or templates," and "explain your obligations... to help manage expectations" [1]. None of these is ICO-provided; each is a project the guidance hands to the organisation that already told the regulator, in the same document, that it is under additional resource pressure. The guidance that names the problem does not fund, template or resource the fix.

The claims, tested

The guidance's own words What its own cited authority or earlier text shows Verdict
A request that is "clearly malicious, abusive or designed to cause disruption, this may support a decision that it's manifestly unfounded or excessive" [1] The guidance's own cited authority states "the use of such language does not automatically make a request manifestly unfounded" and requires concrete markers of intent, grudge or campaign [2] The new guidance's test for refusal is looser than the detailed guidance it links to as its authority
AI-assisted SARs "may also place additional pressures on your resources," and unverified letters of authority need direct contact with the data subject [1] The only practical steps offered are for organisations to write their own training, guidance and templates [1] A named resourcing burden and verification task, with no funded support or detection criteria attached

Credit where due

Most of the guidance holds the line it sets for itself. The repeated insistence that AI assistance does not invalidate a request, does not automatically make it complex, and does not remove an organisation's duty to carry out a reasonable and proportionate search is a genuinely protective stance for ordinary requesters who turn to AI because they are unsure how to phrase a formal letter. The guidance is also honest that fraud risk is real and rising, rather than pretending AI-assisted requests pose no new problem at all, and it correctly limits itself to interpretation rather than inventing new exemptions: "there are no special rules or exemptions that apply specifically to AI-assisted SARs" [1]. That restraint is the right call; the gap is in how the one exception to it was drafted.

Verdict

Four stars, from two documented mistakes. A regulator issuing fast, practical guidance on a fast-moving problem is not, on its own, a fault; AI-assisted SARs are a real and growing category, and most organisations handling one for the first time need exactly this kind of plain-English steer. The problem is that the guidance's single sentence on refusal travels lighter than the detailed guidance it cites, at the one point in the document where getting the threshold wrong means a person loses their statutory right of access, not just extra admin, and that the burdens the guidance itself names come with no funded or provided answer.

The star score counts two documented mistakes. First, the guidance's statement that a "clearly malicious, abusive or designed to cause disruption" request "may support a decision that it's manifestly unfounded or excessive" omits the qualifier and concrete markers set out in the ICO's own detailed guidance on the same test, which states that abusive language alone "does not automatically make a request manifestly unfounded." Second, the guidance names AI-assisted SARs as creating "additional pressures" on organisational resources and a new letter-of-authority verification burden, then offers no ICO-provided support, tool or detection criteria, only a suggestion that organisations write their own training and templates. Two falls in the 1 to 3 band: four stars; the bands are on the ratings page. This piece makes no finding about the ICO's wider AI-and-data-protection work, the accuracy of any specific SAR decision made under this guidance, or whether AI-assisted SARs are in practice being wrongly refused, none of which this piece independently tested. Both pages were fetched and verified live on 14 November 2026.

Sources

  1. Subject access requests generated by artificial intelligence, Information Commissioner's Office, published 11 November 2026
  2. When can we consider a SAR to be manifestly unfounded or excessive?, Information Commissioner's Office, last updated 8 December 2025
SHARE THIS ARTICLEXBlueskyMastodonLinkedInRedditEmail