Characterizing Agentic Flooding of Government Services
A study recently identified 84 potential cases of agentic flooding across 11 different jurisdictions. This happens when AI agents, used by the public to access complex policies or apply for benefits, generate a sudden and massive surge in demand that the government is not prepared to handle. It is a side effect of LLMs making it extremely cheap to automate high-volume interactions.
So, we must identify where the risk is highest. It is not every single service, but specifically those that are financially attractive yet structurally complex. I have spent years watching digital transformation projects in Italy fail because they were built for a human-to-human interaction model that is now being bypassed by automated scripts (and this happens across the EU). If an agent can submit a thousand subsidy requests in a minute, a standard web form will not survive the load.
Now, this puts a massive burden on procurement and governance teams. It is a balancing act. The Public Administration is often slow to react to these shifts, focusing on the front-end interface while the underlying infrastructure remains vulnerable.
But we cannot simply solve this by introducing fees or other friction-inducing measures. Doing so would create a barrier for the citizens who actually need these services to survive, which is the exact opposite of what a public system should do.
What is actually needed is a technical capability to identify automated intent at the very first point of contact, allowing the administration to maintain accessibility without letting the system be overwhelmed by SYNTHETIC demand.
#AI #PublicAdministration #DigitalTransformation #Governance #GovTech