The government 'AI flood' is old demand past a fallen toll booth

A new study that in most headlines reads like a security story is, read carefully, an economics story. Researchers tracked a phenomenon they call "agentic flooding" — AI enabling sudden surges in requests to public services — and documented 84 cases across 11 jurisdictions, from welfare claims and judicial appeals to freedom-of-information requests. In 69% of those cases, government officials themselves blamed AI for the surge. That sounds like an attack. Here is what makes it not an attack: in 87% of the cases, the flood came from ordinary people pasting AI-written text into forms.

No bot nets. No hacked servers. Just applicants, complainants, and litigants who suddenly found the thing they were already owed was now cheap to ask for.

What AI actually removed

The cleanest way to see it is an English number. Complaints to the UK housing ombudsman more than doubled after ChatGPT arrived, rising from about 2,600 in 2022 to over 7,000 last year. US consumer-finance complaints grew fivefold over the same stretch, and German social courts blamed AI for a 55% year-on-year jump in their caseload. What these have in common is that they were never empty gestures — these are people with real claims who now have a free lawyer in their pocket. The study splits the flood into two types, and both are telling: 90% of the cases were "qualitative" flooding (the same or fewer people, but each filing longer and more legally sophisticated) and 60% were "quantitative" (more of them). Half the cases were both.

Read the mechanism underneath and the real finding drops out. The gatekeeper for these services was never genuine scarcity of deserving cases. It was administrative friction — the legal knowledge, the formatting, the psychological cost of wrestling with a bureaucracy. That friction was doing the rationing. It is the toll booth. When an LLM collapses the toll to near zero, the latent demand that was always sitting there simply surfaces. The flood is not new demand; it is old demand that finally got past the gate.

That is the part worth pausing on, because it tells you the surge is structural, not a blip. Most cases showed no sign of slowing. This is a demand curve moving, not a one-off event.

Who is forced to absorb it

Now name the forced actor, because that is where the money angle sits. A government agency cannot ignore a complaint, an appeal, or an FOIA request — there is a statutory clock running against it. So as the case load rises, the processing side of these agencies has to eat the growth on a fixed budget. The cost of the flood does not land on the people who filed; it lands on a balance sheet that is legally barred from saying no.

That constraint is exactly what makes this an investor's question rather than just a think-piece. When the marginal cost of asking collapses but the processing layer is legally bound to absorb every request, a durable and growing share of the value migrates to the plumbing that sits between an angry citizen and a decision. The study already counts one signal of that migration: about a quarter of the surveyed agencies have quietly started using AI tools to process the very request flood that AI helped generate.

So the beneficiary trade is not the shiny "AI conversation" names at the front of the press release. It is the detection-and-absorption layer — the systems that fingerprint anomalous complexity, route it, and clear the statutory backlogs — plus the verification layer. Think about that second one in the crypto register, where the same problem has a name and a decade of practice: the sybil problem. When generating human-grade text is effectively free, the scarce input stops being the words and becomes proof that a real person with a real need is behind them. Identity and proof-of-humanity stop being a compliance checkbox and become an allocative asset. That is a genuinely new economic object — a toll the flood itself created, because the old toll booth is gone.

The trap governments keep reaching for

There is a real risk in the story, and it is not the flood. It is the response to it. Frictionless filing is new, but the instinct it provokes is ancient, and the early responses are blunt instruments: Australia floated bringing back fees on FOI requests to discourage the wave; Japan blocked an entire comment procedure by IP address. Fees, caps, and blocking are exactly the kind of wall that punishes the legitimate, less digitally literate user far more than it stops the determined one. If governments reach for that wall, they are not defending anything — they are rebuilding the toll booth AI just knocked down, and re-entrenching the exact inefficiency that made the flood worth reporting in the first place.

That is the variable to watch, and it is a policy judgment more than a data point. The study itself is careful not to overreach, conceding it cannot prove AI caused every surge even where officials say it did. But the pattern is consistent enough to state plainly: the constraint on these systems was never capacity, it was the cost of asking. That cost is gone for good. The durable winners will be the people who can process and verify the honest flood at scale without recharging the user a toll — and the losers will be the agencies that mistake their own toll booth for security.