You’ve spent hours on hold. You’ve filled out forms that ask the same question three different ways. You’ve been told to fax something to a number that’s been disconnected for years. And every time, you felt it—that quiet, grinding suspicion that the system was designed to wear you down.
That suspicion is correct. And now AI is proving it.
A new paper from researchers studying “agentic flooding” of government services has the tech world buzzing. But almost everyone is reading it wrong. The headlines scream about bots overwhelming public systems, about malicious actors gaming the welfare state, about the need for better CAPTCHAs and digital moats.
They’re missing the real story.
The AI agents aren’t the problem. They’re the diagnostic tool that finally reveals the disease.
Let me say that again: “The ‘flooding’ is actually a diagnostic: it reveals that the system’s real failure was always its reliance on artificial difficulty to manage demand, not the volume of requests.”
Think about it. For decades, government services—from unemployment benefits to food stamps to disability claims—have been deliberately made hard to access. Hard to understand. Hard to complete. Hard to appeal. Why? Because the systems are underfunded. They can’t handle the volume of legitimate demand. So they ration access through complexity.
It’s a silent triage. The people who give up? They never get counted. The people who persist? They become the system’s definition of “deserving.”
I saw this firsthand. Last year, my neighbor—a 62-year-old with a bad back and no computer skills—spent three months trying to apply for disability benefits. She made seven phone calls. She mailed documents that were “lost.” She was told to create an online account, then told her password was wrong, then told to visit an office three towns over. She gave up. She’s still working a job that’s destroying her body because the system was designed to make her quit.
Now imagine an AI agent that can fill out that form in 30 seconds. That can navigate the phone tree. That can resubmit when the portal crashes. That never gets frustrated and never gives up.
Suddenly, the floodgates open. And the system screams, “Attack!”
But it’s not an attack. It’s the first honest conversation we’ve had about capacity in a generation.
“LLMs are just revealing how many systems rely on making applications artificially difficult, either to understand or to generate, to reduce volume. These systems have been failing silently for years—unmuting the alarm is the first step to fixing them.”
That’s from a comment on the original paper. The paper itself is careful—it doesn’t use the word “nefarious.” But the framing matters. The word “flooding” implies a flood is bad. What if the flood is actually the water finally reaching its natural level?
I’m not saying AI agents are perfect. They hallucinate. They can be used for fraud. But so can fax machines, and we don’t call faxing a “flood.”
What we’re really afraid of is the truth: that our public services have been propped up by a system of deliberate exclusion. That the complexity wasn’t a bug—it was the feature. And that AI is now the whistleblower no one wants to hear.
So here’s the choice. We can build better walls. More CAPTCHAs. More identity verification. More layers of friction that will once again filter out the people who can’t afford a lawyer or a bot.
Or we can fund the damn services. Hire more caseworkers. Simplify the forms. Build systems that assume everyone deserves access, not just the most persistent.
“The same complexity designed to ration scarce resources is now being bypassed by AI, forcing a choice between continued underfunding and systemic redesign.”
Don’t blame the bots. Blame the lie that made them necessary.
This affects you. Whether you’re applying for a passport, a small business loan, a building permit, or a healthcare subsidy—you’re navigating a system that was engineered to make you give up. AI just showed us what happens when someone doesn’t.
The question is: will we fix the leak, or will we finally admit the dam was never meant to hold water for everyone?
FAQ
Q: Isn't this paper warning about AI being used for fraud and identity theft against government systems?
A: The paper raises those concerns, but the data shows that what's actually happening is AI agents successfully completing applications that humans were blocked from. The 'fraud' risk exists, but it's dwarfed by the systemic exclusion of legitimate users.
Q: What's the practical implication for someone who needs to use a government service today?
A: If you can afford it, an AI agent can now help you navigate the system—but that creates a two-tier system where the wealthy bypass the friction. The only real fix is to demand that services be redesigned for simplicity, not scarcity.
Q: Isn't the real problem that AI agents will overwhelm already underfunded systems and make them worse for everyone?
A: That's the short-term shock. But the long-term benefit is that the 'overwhelming' is the first honest data we've ever had about true demand. Without it, underfunding stays invisible. With it, we finally have a number that forces accountability.