The scenario that AI policy researchers spent years treating as hypothetical arrived quietly this week. TechCrunch reports that AI agents are flooding public services with new requests, and the twist that every breathless headline buried is this: the vast majority of cases are legitimate. People who were entitled to claim something are now claiming it, because a machine finally made the process frictionless enough to bother. This is not fraud. It is automation finding the gap between what people deserve and what bureaucracy historically made them fight for.
The Friction Was the Policy
This connects directly to what a 2026 paper by Etienne Brown in arXiv frames as the political equality problem of algorithmic reach. Brown argues that the capacity to influence outcomes through speech, or in this case through legal claims, depends not just on having rights but on having the infrastructure to exercise them. Benefits systems were designed with friction baked in: long forms, jargon, deadlines, appeals processes. That friction was never neutral. It disproportionately filtered out claimants with less time, less literacy, and less access to professional help. AI agents are dissolving that filter. The bureaucracy is experiencing something that feels like an attack but is structurally indistinguishable from equity.
What Comes After the Flood
The harder question is what governments do next. The obvious move is to rate-limit or authenticate AI-generated requests. The EU AI Act's Article 50 watermarking obligations, which came into force in August 2026 according to a recent arXiv paper by Nemecek et al., require generative AI outputs to be labeled. But a watermark on a claims form does not tell you whether the claim is valid. It only tells you a machine filed it. Blocking machine-filed legitimate claims in the name of system integrity is a policy choice with real human costs, and governments will have to make it explicitly rather than by default. The flood is the accountability mechanism. The question is whether anyone in a position of power is willing to say so out loud, or whether the response will be to rebuild the friction and call it security.