Guidance with handling AI-generated Freedom of Information requests

The increased accessibility to free, online AI tools in recent years creates the opportunity for individuals to use the technology to quickly draft and submit requests under the Freedom of Information Act 1982 (Cth) (the FOI Act).

This presents novel challenges for Australian government agencies processing Freedom of Information (FOI) requests, who will be required to interpret and apply the FOI Act to requests generated entirely or partly using AI.

To assist government agencies, this article provides guidance in handling AI-generated FOI requests.

Is an FOI request made using AI invalid?

A request is valid and must be processed if it meets all of the requirements contained in s 15(2) of the FOI Act. Namely that the request:

1. is in writing
2. states that it is an application for the purposes of the FOI Act
3. provides adequate information to identify the documents being sought, and
4. provides details about how the applicant can be contacted in relation to the request.

If an FOI request meets all of these requirements, the fact that it may have been drafted entirely or partly using AI does not have any bearing on whether the request is valid or not.

However, if the request is not being made by a ’person’, it is invalid. On this point, paragraph 3.22 of the FOI Guidelines contains the following:

The legally enforceable right of access in s 11 of the FOI Act is conferred on ‘every person’. Section 2C of the Acts Interpretation Act 1901 extends the definition of ‘person’ to include ‘a body politic or corporate as well as an individual’. Without a person or corporate entity making an FOI request (even if assisted by AI), there is no right of access under the FOI Act.

Paragraph 3.22 continues to contemplate the concept of ’bot’ generated FOI requests as being invalid. It is not clear what level of involvement a person must have in the submission of an FOI request for it to be considered a valid request, as opposed to being ’bot’ generated. In any case, agencies should be aware of this distinction when reviewing and processing FOI requests made using AI.

What if the AI-generated request contains hallucinated, incorrect or inaccurate information?

AI tools, particularly generative AI applications, are known to occasionally hallucinate, meaning they can fabricate content such as case law, legislation, sections of legislation, and other legal material. They can also provide inaccurate or incorrect information.

Even if an AI-generated request contains hallucinated material or inaccurate or incorrect information, this does not render it an invalid request. If the request satisfies all the requirements contained in s 15(2), it must be processed as wholly as possible, notwithstanding the hallucinated, inaccurate or incorrect information.

In circumstances where hallucinated, inaccurate or incorrect information may affect the applicant’s expectations in relation to their request, wherever possible, agencies should contact the applicant to correct their understanding and advise them of their rights under the FOI Act. It is important that government agency staff processing FOI requests have a good understanding of the FOI Act in order to identify if an FOI request contains any such error.

How do I handle AI requests that are unclear or overly broad?

The use of generative AI means that applicants can produce lengthy FOI requests with minimal effort. In some cases, the length may make the scope of the request unclear or mean that the scope is far too broad to be processed.

In the first instance, agencies should informally contact the applicant if their request is unclear or is overly broad. Assuming the applicant has used AI to draft the request, asking the applicant to explain the scope in their own words can help to clarify or narrow the scope of the request.

If contacting the applicant informally isn’t successful, agencies can consider formally commencing a request consultation process under s 24AB of the FOI Act, on the basis that a practical refusal reason exists, and attempt to reach a formal resolution with the applicant through this process

Does the use of AI affect whether an applicant is considered vexatious?

Noting that the FOI Act came into law long before generative AI technology became widely available, whether an applicant makes use of AI to make an FOI request is not explicitly mentioned in s 89L of the FOI Act as a relevant consideration for the Information Commissioner when determining whether an applicant is vexatious.

However, the wording of s 89L(1)(a) suggests that agencies may be able to advance arguments that an applicant’s use of AI is disingenuous and contributes to their status as a potentially vexatious applicant.

Section 89L(1)(a) states that:

(1) The Information Commissioner may make a vexatious applicant declaration in relation to a person only if the Information Commissioner is satisfied of any of the following:

 (a) that:

 (i) the person has repeatedly engaged in access actions; and

 (ii) the repeated engagement involves an abuse of the process for the access action;

It is conceivable that an applicant’s use of generative AI to facilitate making a large volume of FOI requests in a short period of time will be a factor supporting a finding that an applicant’s ‘repeated engagement involves an abuse of the process’.

If your agency requires advice or assistance with the processing of FOI requests, please reach out to Chantal Tipene and our team of FOI specialists would be happy to assist.