UK ICO flags the same rights but new practical challenges for AI-assisted data subject access requests
Published on 17 November 2026
Employers face lengthy demands and repeated correspondence while their own technology creates more personal data
At a glance
Length alone does not establish complexity of a request, and clarification remains available where the information sought is unclear.
Requests for metadata raise questions about what counts as personal data and where a proportionate search ends.
Personal data created by employers’ own AI tools may fall outside existing records of how information is held and used.
A data subject access request (DSAR) drafted with the assistance of artificial intelligence (AI) can look daunting: lengthy wording, unfamiliar legal references and an extensive list of systems to search. For the organisation receiving it, the challenge is separating what the individual is actually asking for from the complexity of how the request has been written.
The Information Commission’s Office (ICO) has published new guidance on AI-assisted DSARs. Its starting point is clear: using AI does not invalidate a request or create special grounds for refusing it. The usual rules apply, alongside practical challenges around accuracy, security, deadlines and searches. The task is to distinguish a request that looks complicated from one that is genuinely difficult to answer.
The same underlying rules
AI use neither reduces the individual’s rights nor removes the practical options available to controllers, the ICO explains. Organisations can still seek clarification where reasonably needed, carry out reasonable and proportionate searches, extend the response deadline where justified, and refuse requests that meet the threshold for being manifestly unfounded or excessive. Those decisions depend on the request itself, not the fact that AI helped draft it.
AI use does not automatically make a DSAR complex. Lengthy, unclear or repetitive wording is not enough: organisations should focus on genuine difficulties in locating, retrieving, reviewing or disclosing the information. The usual one-month response deadline applies, with an extension of up to two further months where justified by complexity.
AI use, length, volume or assertive wording alone does not make a DSAR manifestly unfounded or excessive. Clearly malicious, abusive or deliberately disruptive requests may support that conclusion, but organisations will need to assess all the circumstances.
Discussion or clarification may help establish whether a DSAR can be refused on the grounds it is manifestly unfounded or excessive, the ICO says. Where an organisation concludes that these provisions apply, they are required to hold supporting evidence and be able to explain its reasoning to both the individual and the ICO. It recommends keeping a written record of the decision.
The usual scope rules also apply. AI-generated wording may ask for information beyond the right of access, such as details of search parameters, information that is not the individual’s personal data or a detailed explanation of potential exemptions. These additional demands would need to be distinguished from the request for personal data, rather than assumed to define what the response must cover.
Accuracy and searches
An AI-assisted DSAR may contain incorrect legal references, invented case law or inaccurate contact details, the ICO notes. These errors do not invalidate a clear request for personal information.
Organisations remain responsible for the security of personal information, even where the individual supplies incorrect contact details. Where identity or authority checks are needed, organisations should be alert to potentially falsified identity documents or letters of authority. Checking correspondence details against existing records and investigating discrepancies before any disclosure of personal information can help verify identity and authority, rather than treating the provision of a document as conclusive verification.
The same practical approach applies to searches: organisations will need to carry out a reasonable and proportionate search. AI-assisted DSARs typically refer to a more extensive list of sources, such as datasets, backup records, platforms, CCTV footage or call logs. The organisation remains responsible for deciding appropriate search parameters and justifying exclusions.
Relevant search locations are those that are likely to hold the individual's personal information. Other departments, processors or joint controllers may be involved where appropriate. Consultation with the individual before narrowing the scope can avoid assumptions about what they want. Clarification may help where a request is large and, in whole or part, unclear or vague.. A record of what the search includes and excludes, the reasons and any consultation with the individual are required to justify the approach and demonstrate accountability if subsequently challenged.
Metadata requests
Requests for metadata need to be carefully considered, even where they appear AI-assisted, the ICO notes. Metadata may constitute the individual’s personal data if it identifies them directly or indirectly. Common examples include access records, working patterns and document audit logs. An AI-generated list is not a checklist to be automatically followed in full. The relevant questions are whether the information is the individual’s personal data and whether the searches required to locate it are reasonable and proportionate. Extensive, itemised requests may include material of limited value to the individual or fall outside the scope of a reasonable and proportionate search.
Where an itemised request makes excessive demands, explaining that only a reasonable and proportionate search is required can help manage expectations. This is distinct from deciding whether the DSAR itself meets the threshold for refusal as manifestly excessive.
Specifically requested metadata and information reasonably likely to contain relevant personal data will help to determine the search parameters.
Asking for clarification can also establish what the individual means by "metadata", without requiring them to exclude information. Retrieval difficulties can also make the request complex, but this does not automatically justify refusal as manifestly unfounded or excessive.
Making requests easier to handle
Organisations can explain, in neutral language, any apparent misconceptions about data protection law or unnecessary breadth or complexity, potentially arising from how AI helped draft the request. This can help individuals identify the information they want.
A straightforward, accessible process can reduce unnecessary complexity, the ICO observes. Support can help without discouraging people from using AI to exercise their rights.
Clear guidance on making a DSAR and provide a contact point can help individuals frame their requests, while staff training can support the handling of lengthy or complex requests, including checking identity, authority and contact details where needed. Clear, specific requests help identify and locate information, and encourage individuals to check AI-generated wording for accuracy, while optional prompts or templates can make clear that individuals do not have to use them.
Separate ICO guidance
The ICO has also published separate guidance for individuals on using AI to draft data protection requests and complaints, encouraging clear wording, accuracy checks and care with personal information and legal advice. Rather than discouraging AI use, it aims to help individuals use it more effectively, producing clearer DSARs that should ultimately be more manageable for organisations to answer.
Practical challenges may continue throughout the DSAR process and after a response has been provided, particularly where dissatisfied individuals use AI to generate repeated follow-up correspondence. The ICO cautions individuals that AI-generated messages can become repetitive, overly long or more confrontational than intended, and says it may stop responding where contact is unreasonable or persistent. Its approach offers a useful reference point for controllers considering how to set boundaries around similar exchanges.
Osborne Clarke comment
There is another side to this discussion that the guidance does not address: organisations' own use of AI. AI tools generate new categories of personal data from productivity analytics and sentiment scoring to AI-generated HR documentation such as grievance and disciplinary outcomes, Existing records and data maps may not yet capture this information.
Organisations are rightly focused on managing inbound AI-generated DSARs, their own use of AI is also expanding the personal data that must be identified, reviewed and potentially disclosed in response to any DSAR. This makes data mapping more important than ever, with organisations needing to understand what data their AI systems are creating, where it is held and by whom, before they can respond comprehensively to any request.
Appropriate transparency information for the workforce also matters: a DSAR should not be the first time an employee learns how their employer is using AI in relation to their personal data.