2026-11-11OPINION · REGULATORYGOVERNANCE · ADMINISTRATIVELAW · AICOMPLIANCE · PUBLICSECTORACCOUNTABILITY · INFORMATIONGOVERNANCE6 MIN READ READ
FILED UNDER

How AI Complaints Are Breaking Regulatory Bodies

AI-generated correspondence has collapsed the cost of complaining. The obligations to respond have not moved. Something has to give.

The orthodox view inside Australian complaint-handling bodies is that AI-generated correspondence is an operational nuisance: a resourcing problem to be absorbed, then solved with better tooling. That view is wrong. It is a governance problem, and it is already creating legal exposure that most boards and commissioners have not seen, because the decisions generating it are being made by team leaders under pressure rather than by anyone accountable for them.

The UK is a year or two ahead of us. The Guardian reports MPs’ offices, councils, schools and law firms buried under chatbot-generated grievances, requests and legal challenges. One experienced caseworker estimates half of everything her office receives is AI-produced, and says it is longer, more complex and harder to deal with. Every Australian body with a statutory or regulatory duty to respond should read that as a forecast, not foreign news.

Volume used to mean something

Our complaint and review machinery rests on an assumption nobody wrote down. Correspondence volume is a rough proxy for genuine public need, and conscientious processing of every item is both a legal obligation and a signal of institutional trustworthiness.

That assumption held because writing a complaint cost the complainant something: time, effort, the ability to marshal an argument. The cost filtered volume and roughly correlated with seriousness. Generative AI has driven that cost to zero. A single grievance can now arrive as a ten-page submission citing legislation, case law and regulatory guidance, some of it real and some of it invented. It costs the sender minutes and costs the receiver hours.

The OAIC, AFCA, the Commonwealth and state ombudsmen, the ATO, FOI teams across every agency, and the internal dispute resolution functions of every bank, insurer and super fund regulated under ASIC’s RG 271 all operate on timeframes and duties designed for the old cost structure. RG 271 still expects a final response to a standard complaint within 30 calendar days. The FOI Act still runs its statutory clock. Nobody has repriced the obligation for the new world.

The two failure modes

When input costs collapse and output obligations stay fixed, there are only two outcomes.

Response quality degrades across the board. Caseworkers skim. Templated replies proliferate. The serious grievance, the one describing genuine harm, gets the same rushed treatment as the AI-padded one. This is the default path because it requires no decision, and it erodes the trust these bodies exist to maintain.

Triage gets built. Someone decides which correspondence gets full attention and which gets a filter, a form reply or a declined investigation. This is the more rational response, and it is where the liability sits.

Why triage decisions will be litigated

Australian regulators already have discretionary tools. The Privacy Act allows the Commissioner to decline to investigate complaints that are frivolous, vexatious, misconceived or lacking in substance. The FOI Act provides for practical refusal where a request would substantially and unreasonably divert resources, and for vexatious applicant declarations. AFCA’s rules give it grounds to exclude complaints.

None of these tools were designed for a filter that sorts by suspected authorship or by machine-assessed merit at scale. Each is an exercise of administrative discretion. In the public sector, that makes it reviewable. Procedural fairness, relevant and irrelevant considerations, and fettering of discretion are all live grounds of challenge.

Consider what is happening now in operational teams. A unit facing a backlog starts deprioritising submissions that “look AI-generated”. Someone builds a rule that routes long, citation-heavy correspondence to a slower queue. A vendor’s classification tool gets switched on in a pilot. None of these goes near a board or a commission meeting. Each one sets a precedent.

When a dismissed complainant turns out to have had a legitimate grievance, and statistically some will, the first question from their lawyer will be: on what basis was this complaint triaged, who decided that basis, and was it lawful? An undocumented team-level practice is a very poor answer to that question.

The equity problem nobody wants to name

The awkward fact is that a meaningful share of AI-assisted complainants are exactly the people complaint systems exist to protect: people with limited literacy, people writing in a second language, people with disabilities, and people who cannot afford a lawyer and have, quite reasonably, used the free tool that sounds like one.

A triage regime that penalises AI-assisted correspondence as a category will fall disproportionately on those groups. In a regulator, that creates both a discrimination exposure and a legitimacy crisis. In a regulated entity, it is the kind of systemic IDR failure that ASIC and AFCA are positioned to find.

So the correct filter is never “was this written by AI?” It is “does this disclose a matter we are obliged to deal with?” That is harder to build, and it is the only defensible design.

Automating the triage makes it worse

The instinctive fix is to use AI to sort AI. That moves the problem rather than solving it. It moves it into territory Australia has already been burned in.

Robodebt was not primarily a technology failure. It was a governance failure in which automated processing was allowed to substitute for lawful decision-making, and nobody senior asked whether the process could withstand legal scrutiny. An AI triage layer that quietly determines which grievances a regulator investigates carries the same structural risk. The Royal Commission’s findings will be the first thing quoted back at any agency that gets this wrong.

Private organisations are not exempt. From December 2026, the Privacy Act requires entities to disclose in their privacy policies where personal information is used in substantially automated decisions that significantly affect individuals. Automated complaint triage that determines whether a customer’s dispute is properly heard fits that description comfortably. Most organisations have not mapped it.

What boards and commissions should do now

This is not a future problem. The precedents are being set this quarter, by default. Senior leadership’s job is to replace default with decision.

  • Find out what is already happening. Ask your complaints, FOI and IDR functions directly what informal triage, filtering or prioritisation is in use, including vendor tools in pilot. Assume it exists.
  • Bring triage criteria to the accountable level. Any rule that determines whether a complaint receives substantive consideration is a policy decision. Document it, approve it at the appropriate level, and obtain legal advice on how it sits within your statutory discretion.
  • Filter on substance, never on authorship. Make that principle explicit, and test whether your criteria disadvantage protected or vulnerable groups.
  • Measure false negatives. Regularly sample deprioritised and declined correspondence and have experienced staff review it. If you do not know how many legitimate grievances your triage is dropping, you cannot defend it.
  • Keep a human accountable for every decision to decline. Automation can sort. A named person decides.
  • Tell your minister, regulator or portfolio agency what the obligation now costs. Statutory timeframes were set for a different cost of complaint. If they need revisiting, the case should be made openly and with evidence, not met covertly through degraded service.

The bodies that handle this well will be the ones whose senior people decide, on the record, how they will separate signal from noise and why. Everyone else will have that decision made for them by a court, an inquiry or a complainant who was wrongly ignored, and they will be explaining a practice they never knew they had.

Next dossier
The Help Desk Is Your Perimeter Now →
Engage the author
Adam van Vliet is currently taking on briefs.
Brief Adam
Share