Characterizing Agentic Flooding of Government Services

Agentic flooding is already happening: The researchers documented 84 potential cases of AI agents overwhelming government services across 11 jurisdictions, indicating this is not a hypothetical future problem but a present-day challenge that governments are already facing.

LLMs make flooding easier and cheaper: Large language models enable flooding by making it inexpensive to generate text at scale, allowing AI agents to submit applications, requests, or comments to government services in volumes that human staff cannot process.

Complex but financially valuable services face highest near-term risk: Services that are both complicated (requiring significant explanation) and financially attractive (like benefit applications or licensing) are most vulnerable to flooding, as AI agents can easily navigate complexity while targeting high-value outcomes.

Quick-fix solutions create equity problems: While governments have precedent-based tools to stop flooding (such as implementing fees or friction), these fastest-to-deploy responses often restrict access for legitimate users, potentially undermining equitable public service delivery and disadvantaging those who need help most.

Governments should act now with equity-conscious approaches: The paper recommends that governments take near-term preventive actions before flooding becomes widespread, prioritizing solutions that maintain fair access rather than defaulting to barriers that would exclude vulnerable populations from essential services.