Eighty-Four Cases

There is a number in British public policy that has been quietly load-bearing for decades. In 2025/26, an estimated 24.1 billion pounds of income-related benefits and social tariffs will go unclaimed across Great Britain. Universal Credit alone accounts for 11.1 billion of it. Council tax support, the most underclaimed benefit in England, leaves around 2.8 billion with the Treasury that Parliament voted to give away.

The standard reading of that number is a failure of outreach. People do not know what they are owed, the forms are hard, the stigma is real, and the state should do better at telling them.

There is a second reading, and it has become urgent this year. The system's finances assume that number stays roughly where it is. Not as policy. As arithmetic.

A dataset, and a name for it

On 17 August, Chris Schmitz, Lewis Hammond and Alan Chan published a paper characterising what they call agentic flooding: surges in the volume or complexity of requests reaching government services, driven by the collapse in what it costs a member of the public to produce a competent one. They assembled 84 potential cases across 11 jurisdictions.

Their risk matrix is the useful part, and its conclusion is narrower than the headline. Near-term exposure is highest not for the largest services or the most digitised ones, but for those that are financially attractive and procedurally complex. Money at the end, and enough process in the middle that a meaningful share of eligible people historically abandoned the attempt.

That is a precise description of the machinery that produces the 24.1 billion pound figure. Complexity was never a deliberate rationing device. It has been functioning as one.

The tribunals went first because the incentive was cleanest

Britain's employment tribunals are the clearest instance on the record. Each tribunal historically received in the region of 20 applications for interim relief a year. They now receive that many in a month. Single claims are up 39 percent year on year while disposals have fallen 12 percent, the backlog of single claims has grown by 55 percent, and hearings are being listed into 2030.

On 23 June, Judge Barry Clarke, President of the Employment Tribunal in England and Wales, and his Scottish counterpart issued guidance restating that an interim relief application requires a "pretty good chance of success", a standard deliberately higher than more-likely-than-not. The judicial view, reported alongside the guidance, is that AI assistance is behind the rise in volume and in the complexity of what arrives.

Note what the tribunals did not do. They did not add capacity, and they did not attempt to detect machine-drafted submissions. They restated the standard, in public, in advance, and explained the cost of the alternative: interim relief applications are listed quickly because they are urgent, and to make way for them other hearings are postponed. The queue was already zero-sum, and the guidance said so out loud.

The same tool points both ways

It would be convenient if this were a story about claimants against institutions. It is not. In the English planning system the same capability is being used to obstruct.

Objector.ai sells a policy-cited objection to a full planning application for 45 pounds, and a 249 pound tier for residents pooling funds against a larger scheme. The documents cite policy and engage with the officer's report, which means a planning officer must read and answer them at the same cost as one drafted by a solicitor. Determination times now average beyond 40 weeks against a 13 week statutory target, and grants of permission have fallen to their lowest level since records began more than two decades ago.

Geoff Keal, chief executive of TerraQuest, which operates the national planning portal handling roughly 95 percent of UK applications, describes objectors "using AI to be able to provide better objection documents, much wider and much broader, which is slowing the system down." It should be said that the causal claim there is asserted rather than evidenced; the planning system has several concurrent problems and no published study yet separates them. But the pricing is real, the tooling is real, and the direction is not in dispute.

So the flood is not ideological. It is available to whoever wants to use a process, for or against, and it arrives first wherever the payoff is highest relative to the effort that used to be required.

The uncomfortable version

Strip out the abuse cases and the picture does not improve, which is what makes this genuinely difficult rather than merely expensive.

A person who is owed council tax support and does not claim it is not committing a fraud by claiming it. An employee with a protected disclosure claim who could not previously have drafted an interim relief application is not gaming anything by filing one. If agentic tools do nothing except deliver to people exactly what statute already entitles them to, the resulting volume is not a pathology. It is the system finally being asked to do what it says it does.

It cannot. Not at that volume, not at current staffing, and not with intake processes designed around post and telephone. Which means a decision that has been deferred for thirty years by the friction of the forms is about to be taken explicitly: either the entitlement is real and must be funded at full uptake, or the standard has to be raised and somebody has to say so in public.

The tribunals chose the second and were straightforward about it. That is more honest than the alternative, which is to leave the entitlement nominally intact and let the queue do the rationing invisibly, at 2030 listing dates, on whoever can least afford to wait four years.

What a serious response looks like

Detection is the answer that will be reached for first and it is the weakest one. The submissions causing the problem are not defective. Judges are not objecting to hallucinated citations, though those exist; they are objecting to volume and complexity in filings that are, on their face, arguable. A detector that worked perfectly would sort submissions by authorship and not at all by merit.

Three responses have more in them. Publish the standard before the surge, as the Presidents did, so that a threshold reads as a rule rather than as a brush-off invented under load. Instrument arrivals rather than outcomes, because a step change in volume is visible weeks before it becomes a service failure and most agencies do not hold the series that would show it. And cost the entitlement at full uptake, so that the gap between what is promised and what is funded is a number somebody has written down rather than a discovery made in a select committee.

None of that requires a view on whether AI is good. It requires accepting that a filter nobody designed, nobody voted for and nobody wrote down has been removed, and that a great deal of public administration was resting on it.

Sources

• Chris Schmitz, Lewis Hammond and Alan Chan, "Characterizing Agentic Flooding of Government Services", arXiv, 17 August 2026. Source for the term, the dataset of 84 potential cases across 11 jurisdictions, and the finding on financially attractive but complex services.
• Policy in Practice, "Missing Out". Source for unclaimed income-related benefits and social tariffs across Great Britain, the Universal Credit and council tax support components. The Economist's issue of 8 to 14 August 2026 gives the figure as 20 billion pounds; the more recent Policy in Practice estimate for 2025/26 is 24.1 billion, and we have used the latter.
• Freeths, "Employment Tribunal statistics: A growing backlog and the rise of interim relief applications", 26 June 2026. Source for the claim, disposal and backlog percentages, the 2030 listings, and the per-tribunal characterisation of interim relief volume.
• The Law Society Gazette, "Tribunal tries to deter interim relief bids amid AI-related surge", 23 June 2026. Source for the guidance, its authors, the "pretty good chance of success" standard and the postponement point. The Gazette renders the interim relief increase as a Britain-wide total where Freeths renders it per tribunal; the discrepancy is unresolved in the public reporting and is flagged rather than smoothed.
• Business Matters, "AI-Powered Nimbyism", 20 May 2026. Source for Objector.ai pricing, determination times against the statutory target, permission grants at a record low, and the Geoff Keal quotation. Used for those facts and not for its causal claim, which it does not evidence.