The Information Commission’s Office has published guidance for organisations on how to handle data subject access requests (“DSARs”) that have been generated by, or drafted with the assistance of, artificial intelligence (“AI”).
The key practical points are that organisations:
- should not assume an AI-generated request is always invalid, complex or easier to refuse;
- should continue to apply the UK GDPR rules on DSARs to all DSARs, including AI-assisted DSARs;
- are encouraged to discuss the DSAR with the requester in certain circumstances, for example to explain why a request appears to include misunderstandings about how UK data protection law works due to the way the request was generated, or to explain what information the individual is and is not entitled to under data protection law.
What is important is the substance of the request and whether the person is asking for their personal data, rather than how the request was produced.
An AI-assisted DSAR is not automatically invalid or complex, the normal rules and expectations on DSARs apply
If a data subject uses AI to generate or draft a DSAR, that on its own does not make the request invalid.
If a data subject makes clear they want access to their personal data, the request should be assessed in the usual way.
The ICO states that organisations cannot discourage data subjects from using AI, but organisations can offer to help them with their request.
- The existing rules apply: The ICO is clear that organisations must treat AI-assisted DSARs in the same way as any other DSAR. Therefore, the rules and process for responding to a DSAR under UK data protection law apply to AI-assisted or generated DSARs.
- Incorrect information or references: AI hallucinations, such as inaccurate or false information or references, do not invalidate a DSAR. A DSAR does not need to follow a prescribed format or use particular (or even correct) legal terminology.
- Complex requests and timeframe: The normal rules on timeframes for responding DSARs under UK data protection law apply to AI-generated or assisted DSARs. An AI-generated or assisted DSAR does not automatically make the DSAR a “complex” DSAR. Therefore, the fact a DSAR has been generated or drafted using AI does not automatically mean the timeframe for responding can be extended as standard policy, and each DSAR should be assessed on a case by case basis.
- Clarification: As with any other DSAR, it is permissible to ask the data subject to clarify their request if it is unclear what they are asking for. However, the ICO warns organisations should only ask for clarification if it is reasonable required.
- Reasonable and proportionate search: The requirement to conduct a reasonable and proportionate search for the requested information also applies to AI-assisted DSARs. Although an AI-assisted DSAR may include specific instructions about searches the data subject expects to be conducted, it is the controller’s responsibility to decide the parameters of a reasonable and proportionate search. It is reasonable to confine the search to areas where information is likely to be held and exclude locations where it is unlikely to be held. However, controllers should not narrow the scope of the DSAR, limit searches or make assumptions about what the data subject wants without consulting them.
Metadata
Organisations must consider requests for specific information such as metadata, even if the person used AI to draft the DSAR or does not fully understand what they are requesting. This means organisations must consider whether the requested metadata is the requester’s personal data and carry out a reasonable and proportionate search.
However, if a requester specifically asks for metadata but is not clear what they mean, the ICO indicates organisations could ask the requester to clarify what they mean and could offer to discuss the request with the requester to help them frame it more clearly. Although the ICO indicates organisations could ask the data subject for clarification, the ICO also states organisations cannot make the data subject rewrite their DSAR to exclude information.
A broad request for “all of my personal information” or “all of my personal information, including metadata” does not necessarily mean an organisation must search for every category of metadata or system-generated audit data, but this depends on the circumstances. The principle of a reasonable and proportionate search applies.
Manifestly unfounded or excessive?
The ICO is clear there are no special rules or exemptions that apply specifically to AI-assisted DSARs, and any of the exemptions could apply as they would to any other DSAR.
The use of AI, formulaic wording or inaccurate legal references does not, by itself, make a request manifestly unfounded or excessive.
An AI-generated or assisted DSAR may use a formal tone or assertive language, but this “doesn’t necessarily mean they’re trying to be disruptive or that they don’t genuinely want to access their information”. The ICO encourages organisations where appropriate to discuss the DSAR with the person or ask them to clarify what information they are seeking, before determining whether the request is manifestly unfounded or excessive.
A requester is responsible for the tone of the request, including if AI is used. If a DSAR is “clearly malicious, abusive or designed to cause disruption, this may support a decision that it’s manifestly unfounded or excessive” and this is the case whether or not AI has been used in drafting the request.
If an organisation seeks to rely on the manifestly unfounded or excessive provisions, the ICO states the organisation must be able to justify the decision.
Is the information requested within the scope of the right of access?
AI-generated DSARs may include requests that are outside the scope of the right of access, such as information that does not relate to them.
While a DSAR gives individuals access to their personal data and related supplementary information, it does not necessarily require an organisation to answer every general question about its business, systems or legal compliance. That distinction becomes particularly important where an AI-generated request combines a valid access request with comprehensive questionnaires or demands for explanations that go beyond the right of access.
Organisations are not required to respond to such elements, but the ICO recommends organisations explain to individuals which parts of their request fall outside the scope of what they are entitled to under UK data protection laws.
As AI-assisted or generated DSARs may contain inaccurate references or information, the ICO does not expect organisations to spend time “deciphering legal references or other details that are inaccurate or irrelevant”. The ICO recommends explaining to the individual that “their request appears to include misunderstandings about how data protection law works or is broader or more complex than necessary, possibly due to the way it was generated.” In addition, where appropriate, the ICO states “you can explain what the law does and doesn’t require”.
What about fraudulent DSARs?
The ICO warns organisations of the risk of inadvertent disclosure of personal data in response to an AI-generated DSAR by a fraudulent actor. The ICO states that “fraudulent actors may use AI to deliberately create false information to obtain someone’s personal information – for example, by creating false ID or letters of authority”.
The ICO recommends that if an organisation is unsure of a person’s identity or thinks their identity or authority documents are not genuine, the organisation should verify the person’s identity and request with the person directly.
Practical steps
In the guidance the ICO recommends practical steps for organisations when dealing with AI-assisted DSARs which include:
- Providing data subjects with clear guidance on how to make a DSAR and a point of contact for data subjects who may want advice or help;
- Ensuring staff are trained so they can understand what is required under UK data protection law and are not intimidated by long or complex DSARs;
- Informing data subjects that clear specific requests make it easier for the organisation to identify and locate the requested personal data;
- Publishing optional example prompts or templates that data subjects could use, although if organisations decide to do this the ICO expects them to make it clear to data subjects they are not required to use them;
- Reminding data subjects to review any AI-assisted DSAR before sending it. This could include reminding data subjects to check the correspondence address is correct and to remove irrelevant or unclear details; and
- Explaining the organisation’s obligations under UK data protection law in relation to DSARs, including the timeframes for responding to DSARs, which can help manage data subject expectations.