Citizens’ AI agents alert

Computer Weekly warned on April 14 that citizens’ own artificial intelligence agents could flood public services with automated requests, even when each user is acting legitimately. (computerweekly.com) The premise is simple: a personal software agent can fill forms, check eligibility, compare options, and retry failed requests far faster than a person clicking through a website. Computer Weekly argued that this could turn one citizen into a stream of machine traffic hitting public systems at scale. (computerweekly.com) That warning lands in a government estate that already runs on application programming interfaces, the machine-to-machine links behind websites and apps. The United Kingdom’s API catalogue lists interfaces across departments including HM Revenue and Customs, the Department for Work and Pensions, the Home Office, National Health Service Digital, and Companies House. (api.gov.uk) Public bodies already publish rules for bots and heavy automated use. The Office for National Statistics says its sites and application programming interfaces are limited to 120 requests per 10 seconds, 200 requests per minute, and 15 requests per 10 seconds for high-demand assets, with blocks of up to one hour if clients ignore the `Retry-After` header. (developer.ons.gov.uk) Companies House applies a different cap: 600 requests in a five-minute period per application, then returns `429 Too Many Requests` for the rest of that window. It also says it can ban applications that regularly exceed or try to bypass those limits. (developer-specs.company-information.service.gov.uk) Government guidance already assumes clients will fail and try again. National Health Service England Digital tells developers to handle rate limits and temporary service problems with retry logic such as exponential backoff, which spaces out repeat attempts at 1 second, 2 seconds, 4 seconds, and so on. (digital.nhs.uk) The pressure point is that artificial intelligence agents can follow those rules and still create far more demand than a human user. New America wrote on February 11 that artificial intelligence lowers the friction of applying for benefits, filing complaints, and requesting help, which makes previously hidden demand visible inside government systems. (newamerica.org) Government Digital Service guidance reflects the same operational trade-offs. It tells teams to set service levels for latency, throughput, and availability, and to use tools such as caching and content delivery networks, while warning that sensitive or user-specific data should never be cached. (gov.uk) Standards bodies are starting to prepare for the agent era as well. The National Institute of Standards and Technology created its Artificial Intelligence Agent Standards Initiative in February 2026 and said it will work on agent authentication, identity infrastructure, and secure interactions between humans and agents. (nist.gov) The immediate question for public services is no longer only whether officials will deploy artificial intelligence inside government. It is whether the systems citizens already use can absorb millions of polite, persistent machine clients without locking out the people those services were built to serve. (computerweekly.com)