agentic ai in public sector: The Competition and Markets Authority (CMA) says breaches involving autonomous AI agents could attract fines of up to 10% of global annual turnover, a penalty that makes the procurement process a matter of risk management as well as cost‑saving.
Public‑sector buyers are therefore being told to compare the agentic ai in public sector on method and results, not merely on the name on the brochure.
Why the focus on method and outcome?
Elsewhen’s report notes that the real impact of agentic AI lies in “organisational workflows, agents operating behind the scenes to orchestrate tasks, coordinate systems, and carry out complex operational work that used to take hours of manual labour.” The firm argues that these systems can free civil servants for “judgment, empathy, and strategy.” While the piece celebrates the efficiency gains, it does not address how a department should evaluate which system delivers the promised speed or accuracy.
That gap is what regulators are trying to fill. The ICO’s Tech Futures paper flags “data‑protection risks and opportunities” that organisations must weigh when deploying agentic AI. It stresses that understanding capabilities and associated risks is essential, but stops short of prescribing a procurement checklist.
Regulatory pressure mounting
ReedSmith highlights a recent surge in regulatory activity. On 9 March 2026 the CMA published a research paper that “recognises the potential benefits of AI agents … but flags several risks that the CMA considers heightened by the autonomous nature of these systems.” The paper makes clear that existing UK consumer‑protection law, including the CMA’s enforcement powers under the Digital Markets, Competition and Consumers Act 2024, will apply regardless of whether conduct is carried out by a human or an agentic AI.
Potential fines of up to 10% of global turnover mean that a mis‑step in procurement – for example, selecting a system solely because the vendor’s brochure touts “hundred planning records per day” – could become a costly compliance issue.
What public bodies can do today
- Map the specific tasks the AI will automate and set measurable performance targets (e.g., processing time per record, error rate).
- Request pilot‑phase results from vendors, not just marketing decks.
- Cross‑check data‑protection assessments against the ICO’s scenario analysis.
- Document the decision‑making process to demonstrate compliance with the CMA’s enforcement framework.
These steps echo the advice in GOV.UK’s guidance on agentic AI and consumers, which outlines the need for transparency and accountability when public services embed AI into decision‑making.
Who is affected and when
Departments ranging from education to health are already piloting agentic agents. The Elsewhen excerpt cites “a hundred planning records per day” as a benchmark that could replace the historic average of five. If a local authority adopts such a system without a robust comparison of methods, it may face scrutiny under the CMA’s upcoming enforcement timetable, expected to roll out in Q1 2027.
For civil servants, the shift means a new skill set: overseeing AI‑driven workflows, interpreting performance dashboards, and ensuring that any deviation from promised outcomes is flagged early.
What remains unknown
Neither the CMA paper nor the ICO report spells out a standardised procurement framework for agentic AI. The claim that “buyers must compare method and results” is therefore a best‑practice recommendation rather than a regulatory requirement at this stage. Public‑sector buyers will need to watch for forthcoming guidance from the CMA and the UK government.
In the meantime, departments can learn from early adopters. The AI regulation update on City AM Canada details how one municipality documented pilot results and avoided a potential fine by renegotiating its contract based on verified throughput gains.
Another internal case study, Public‑sector AI pilots, shows that transparent benchmarking reduced implementation risk and built stakeholder confidence.
Until a formal procurement standard emerges, the safest route for public‑sector buyers is to treat the brochure as a starting point, not the final verdict.