GCRI · Artificial Intelligence and Autonomous Systems
AI and public services
Governments use algorithms to target benefits, detect fraud, triage health care, allocate inspections, translate and answer citizens’ questions. When these systems are wrong, opaque or biased, people lose services and rights they depend on, often without knowing why. This page sets out how public sector AI succeeds and fails, how it is assessed and made accountable, and what evidence serves the agencies, auditors and citizens involved.
How the work runs.
Model registry
AI and public services · Model registry · Release gate
Models
| M-843 triage assistant | Public service | High | 22 of 24 suites | Blocked |
| M-266 flood nowcast | Science | Medium | 11 of 12 | Released |
| M-756 document search | Internal | Low | 8 of 8 | Released |
| M-769 agent planner | Agentic | High | 12 of 16 | Evaluating |
Model card
- Intended use
- triage support, human decides
- Out of scope
- eligibility decisions
- Known limits
- lower accuracy on rare cases
- Owner
- Model owner
Incident log
- D-13INCHallucinated citation in summary
- D-6MITRetrieval grounding added
- D-2INCPrompt injection via uploaded file
- D-1MITTool permission narrowed
Evaluation scores
vs release threshold
Fig.Model registry: example names and illustrative values, not a record or a live system.
AI and public services: how the risk develops.
Public accountability, access and rights. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Errors fall on the vulnerable
People who depend most on public services are often those most affected by automated decisions, and least able to challenge them.
Historical data carry bias
Models trained on past decisions can reproduce discrimination present in those decisions, even when protected characteristics are excluded.
Opacity blocks challenge
When people cannot see that an algorithm was used or why it produced a result, they cannot contest it, and errors persist.
Accountability stays with the agency
Procuring a system from a vendor does not transfer responsibility. Agencies must be able to explain, audit and override algorithmic decisions.
How AI and public services is modelled, measured and tested.
The model, its assumptions, its data and its uncertainty, in the form a scientist or engineer would check. Structures follow published methods. The values drawn are illustrative, not findings.
System model
The couplings between the systems in AI and public services: signs of influence, a reinforcing and a balancing loop, and cross-links where stress jumps.
AI in public services touches the domains people depend on.
Public service AI studies are designed for the Evidence Brief series. The EU AI Act classifies many public-service uses as high-risk; UNESCO’s Recommendation (2021) and WHO guidance (2021) set ethical expectations.
- Algorithmic impact assessments a body publishes.
- Accounts from people affected by automated decisions.
What is measured, and how well.
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| Public algorithm registers | count of systems registered | per update | Government algorithm registers | Published statistics |
| Algorithmic impact assessments completed | percent of deployed systems | annual | Agency records | DECLARED |
| Error rate by population group | difference in error rates | per audit | Algorithm audits | OBSERVED |
| Decisions appealed and overturned | count and percent | annual | Appeals records | Published statistics |
| Systems with human review of adverse decisions | percent of systems | per review | Agency documentation | DECLARED |
| Audit findings on public sector algorithms | count by severity | per audit cycle | Audit office reports | REPORTED |
Where the evidence comes from.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| Government algorithm and AI registers, such as the UK Algorithmic Transparency Recording Standard records | Public records of algorithmic tools in use | Open, Open Government Licence where applicable |
| National audit office reports on algorithms | Independent audits of public sector AI | Public records, licence per office |
| OECD.AI Policy Observatory | Government AI policies and the AI Incidents Monitor | Public, OECD terms |
| Court and ombudsman decisions on automated decision-making | Legal findings on public sector algorithms | Public records |
| Canada Directive on Automated Decision-Making and its impact assessment tool | A model for impact assessment requirements | Public, Government of Canada terms |
Who decides, and what the evidence serves.
Artificial Intelligence and Autonomous Systems prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Government agencies | Whether and how to deploy an algorithm in a public service | Impact assessments; audit methods |
| Auditors and ombudspersons | Where to investigate algorithmic harm | Registers; appeals data; audit evidence |
| Legislatures | What transparency and accountability to require | Comparative policy and case evidence |
Outputs this topic produces
- Impact assessment templates for public sector AI
- Audit methods for disaggregated error and fairness
- Transparency record templates
- Case studies of public sector algorithm failures and fixes
What public-sector technologists ask.
- Is the system accurate for the whole population it serves, including minority groups?
- Can affected people understand a decision, challenge it and get it corrected?
- Who is accountable when an automated decision is wrong?
- Does automation improve access, or does it create new barriers for people without digital skills or connections?
How automated decisions go wrong.
Systems trained on historical administrative data inherit its biases and gaps. Risk-scoring tools can concentrate scrutiny on groups already over-represented in past enforcement. Small error rates become large numbers of wrongly affected people at national scale. When staff defer to the system, known as automation bias, errors are not caught.
Opacity compounds harm: if neither public servants nor citizens can see why a decision was made, it is hard to contest and slow to correct. Access barriers arise when services move online without alternatives.
Evidence for accountable systems.
- 01
Impact assessment
Structured assessment of a system’s purpose, data, affected groups and safeguards before deployment, using social and behavioural science. Quality depends on honest completion.
- 02
Fairness and error analysis
Disaggregated error rates by group, with probability and statistics. Different fairness definitions can conflict, so trade-offs must be made explicit.
- 03
Explanation and contestability
Designing decision notices and appeal routes people can use. Technical explanations do not by themselves make a decision contestable.
- 04
Audit and monitoring
Ongoing checks of outcomes and complaints after deployment.
How it is studied, and what the methods cannot say.
- 01Social and behavioural scienceHuman oversight studies · Affected community research · Workplace adoption studiesApplied here
- 02AI evaluation and assurance engineeringAssurance cases for AI systems · Structured red teaming · Post-deployment monitoringApplied here
- 03Probability and statisticsUncertainty on benchmark scores · Calibration assessment · Audit samplingApplied here
- 04Artificial intelligence and machine learningDistribution shift analysis · Fine-tuning for a specific use · Interpretability methodsApplied here
Where this page’s references lead.
The bodies and documents this page is read against. Status and editions change; each link goes to the issuing body.
Questions and answers.
What is automated decision-making?
What is an algorithmic impact assessment?
Can people challenge an algorithmic decision?
Should public services use AI at all?
What is automation bias?
Can affected people use the platform’s work?
Does the platform evaluate specific government systems?
Contribute evidence from this page.
Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
- EmailGateway GW-GCRISend it from this pageEvidence, a correction, a case, a method, a review, a translation or profile input. It arrives tagged with this page, so the editor who owns it receives it.
- ZenodoOpen depositDeposit it and submit to the communityPublish your own record on Zenodo with your ORCID iD, then submit it to the Reports community there. Curators review it; the DOI stays yours.
- ReportsGCRISee every way to contributeThe contribution types, how each is reviewed, what credit you receive, the licence and what is never accepted.
Never send personal data, sensitive infrastructure detail or anything you have no right to share. A contribution is reviewed before use; sending it confers no authorship or editorial rights, and sponsorship never buys a finding.
Bring a method, a dataset or a question.
Information and enquiries are open today. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.
Legal documents.
Maps, data and figures on this site are shown for context and general information. They are not advice, forecasts or warnings; the limits of the information are set out in the Disclaimers and Data Notices.