How can government prepare for agentic flooding?

For all the progress made in digital government and service design, interactions with government are not always easy. Many struggle to determine their eligibility for entitlements or how to access them, meaning many never receive the services or support available to them. Yet that friction also had an unintended constraining effect: the effort required to engage with government limited both demand and the volume of material submitted.

These barriers are now vanishing. AI is already lowering the effort required to interact with government. It can help people understand what they may be eligible for, navigate complex processes, and prepare applications or submissions.

The consequence, already playing out across the public sector, is a dramatic increase in the volume and complexity of public submissions. In a recent survey of 512 frontline civil servants, more than nine in ten reported that AI-generated content from the public was already affecting their work to some degree, with 58 percent describing it as a significant issue.

Whether it’s FOI requests or complaints about water companies, local authorities, or the NHS, anything traditionally involving paperwork is seeing a sharp spike in public participation. The single-claim caseload at Britain’s employment tribunals rose 39 percent in the year to March 2026, while resolutions fell. One historically obscure emergency provision, ‘interim relief’, which lets a dismissed employee ask a tribunal to force reinstatement or continued pay, has gone from roughly 20 applications a year nationally to over a hundred times that. As for the length of submissions, one reported case involved an NHS employee’s AI-drafted claim running to 282 pages and 67 separate allegations. There are also reports of AI-assisted claims misstating the law or drawing on cases from other jurisdictions.

These pressures are only likely to intensify. AI agents can increasingly string together multiple tasks towards a goal with greater independence, from identifying relevant services to assembling information and completing parts of a process on someone’s behalf. Rather than simply helping a person make one submission, they could enable interactions with government at a scale and frequency that would previously have required significant human effort. With permission to access accounts, retain credentials and interact with services, always-on agents could monitor for opportunities and initiate or pursue actions without waiting for a person to start each interaction.

Already, the chair of the Public Accounts Committee has warned that AI is “approaching government like a tsunami”.

So, what can government departments and service providers do now to prepare themselves for machine-scale demand?

Understand your exposure

There are some ‘no regret’ decisions that public sector leaders can take today. One is to take the time to systematically understand the threat faced.

In many cases, the areas most exposed will be obvious. The harder question is how severe that exposure could become. It may not be possible to predict AI-driven demand precisely, but intelligently aggregating existing data will make it easier to anticipate the scale of the influx.

Through careful scenario planning, ask how much pressure this could put on teams. If agents flood your systems with requests, what would that do to service provision? Would the service be as equitable, or would those without access to AI be at a disadvantage? How would you respond, and where would you prioritise, if this happened tomorrow?

Plans should test how services cope as AI moves from assistance to being able to string together multiple tasks towards a goal with greater independence. A service that copes with AI-drafted letters may not cope when agents start submitting on people’s behalf or acting for them continuously in the background.

Leaders will need to assess the impact across the whole service, not just where AI-generated demand first appears. Changes that relieve pressure in one area could simply displace it elsewhere or degrade the service overall. Does the organisation have the necessary data, technology, service-design expertise and operational capacity to respond across the system?

And some responses may go beyond service design. Where aspects of a service are prescribed in regulation or law, even a well-designed and well-resourced service may be unable to change without regulatory reform or primary or secondary legislation.

Avoid an AI arms race

In response, some organisations are exploring how AI-enabled case handling could support staff in managing growing volumes of material. We supported HM Courts and Tribunals Service (HMCTS) in exploring AI use cases to improve productivity in the courts. Given the ethical and legal sensitivity of what HMCTS does, it was essential that any AI outputs remained unbiased and accurate, while respecting the boundaries of judicial decision making. We worked with them on a risk-based approach to AI adoption that resulted in them starting with useful tools for back-office tasks like knowledge retrieval, rather than pushing for frontline applications.

Examples such as these demonstrate how AI can help public bodies manage growing volumes of work. Other organisations are exploring more defensive measures: online forms redesigned with word limits so that only relevant information is provided; site access denied to agents that do not identify themselves or follow a service’s rules; and agents prevented from autonomously submitting applications.

These are sensible short-term measures but are unlikely to last. Word limits help with AI-drafted text but do little against an agent signed in as the citizen. And simply blocking agents risks shutting out a technology that could become a primary way people interact with digital services. Rather than continually adding new barriers, government should ask how agent-mediated access can work safely and efficiently. That does not mean accepting all demand indiscriminately: services will still need safeguards against fraudulent, abusive, or deliberately distortive activity.

So, with agents likely to mediate much of the public’s future interaction with public services, what should the response be? The deeper change is a wholesale realignment of how people interact with government services, one that could make services easier to access while reducing the cost of providing them.

Redesign the system, not just the interface

Consequently, organisations should start preparing now for a more fundamental redesign of how citizens, their agents and public services interact.

At the most basic level, public sector organisations need to ensure that the requests that land in a queue today are captured as structured data to which data science and analytics can be applied. That’s the point from which leaders can start looking at long-term, structural responses.

Over the longer term, government should create a dedicated route for agentic interaction: programming interfaces (APIs), Model Context Protocol (MCP) tools, and even novel standards like auth.md that allow AI agents to identify themselves and submit clean, structured data rather than sprawling documents.

Each agent should have its own identity, so it never signs in as the citizen. Citizens could use GOV.UK One Login to grant agents permission for specific services and actions, with access determined by the context, limited in scope and time, and revocable. Services could then verify the agent and its permissions on every request, accept structured data, and pre-fill information government already holds.

The principle is that agents should be able to do the administrative work without assuming the citizen’s authority. Legally significant actions, such as declarations of truth, would return to the person for confirmation, while citizens could review what an agent had done on their behalf.

The infrastructure to support this model does not yet exist. GOV.UK One Login does not currently support acting on someone else’s behalf, and the UK digital verification trust framework covers human representatives, not software. Government needs to close that gap. Open Banking points to a potential model, with regulated third parties using APIs with customer consent. If government gives legitimate agents a faster, authorised way to interact with services, they have less reason to scrape web forms or impersonate citizens. Legitimate agent activity would then pass through a recognised route, making activity outside it easier to identify and investigate as potentially fraudulent.

Use AI to widen access

Preparing for machine-scale demand is essential. But it would be a mistake to view citizen use of AI solely as a risk to be managed.

At present, there are real barriers to this access: Policy in Practice’s most recent research estimates that £24bn of income-related benefits and social tariffs goes unclaimed each year across Great Britain.

The cost and complexity of accessing government services not only makes them less effective in principle; it is also inherently inequitable and drags on growth. If someone is deterred from accessing support in getting back to work or adjusting to a new disability, for example, everybody loses.

This aligns with the government’s own blueprint for modern digital government, which envisages services working across institutional boundaries to provide a more timesaving, personalised user experience, while reducing the bureaucratic burden on the public.

Government has already made major strides in digitising important public services, but there is still more to do. The 2025 State of Digital Government Review found that 47 percent of central government services lacked a digital pathway. AI could help accelerate that by connecting data, automating routine processing, enabling greater self-service, and reducing the amount of information that citizens and staff must submit repeatedly.

Done well, this could make services easier to use and more efficient for everyone. We should not preserve existing ways of working when they stand in the way of that opportunity.

Government’s response therefore needs to work at three levels: understand where machine-scale demand will place services under greatest pressure; manage the immediate influx without entering a permanent cycle of barriers and workarounds; and redesign services for secure, structured interaction between citizens, agents, and the state. Throughout, the objective should not simply be to suppress demand, but to manage it intelligently while making legitimate access more sustainable and equitable.