AI, Public Services and Accountability: Who Is Responsible When the System Gets It Wrong?

The AI Funding Files: Following the Money, Examining the Impact

Artificial intelligence is increasingly being explored for public services, from helping staff process information to supporting decisions that affect people's benefits, housing, education and access to care. Used carefully, it could help organisations manage workloads and improve services. Used poorly, it could make mistakes harder to identify, challenge or correct.

When a public body uses an AI tool, who is responsible for its output? Can the person affected find out that the tool was used? Can they understand how it influenced the decision? And if something goes wrong, is there a meaningful route to challenge it?

These questions matter because public services make decisions that can have serious consequences for people's lives. Technology may assist a decision, but it should not make responsibility disappear.

1. Where is AI being used in public services?

AI and other algorithmic tools can be used in different ways across public services. Some help staff organise information or identify patterns. Others may support the assessment of applications, prioritise cases, assist with communications or provide recommendations to decision-makers.

Examples of potential uses include:

• Administrative work: summarising documents, sorting correspondence and helping staff handle routine enquiries.
• Public information: assisting with search tools, website content and responses to common questions.
• Case management: helping staff organise workloads or identify cases that may require attention.
• Planning and analysis: examining data to support decisions about services and resources.
• Decision support: providing recommendations or assessments that a public employee may consider when reaching a decision.

These uses are not all equivalent. A tool that helps a member of staff summarise a public document presents different risks from a system that influences whether someone receives financial support or access to a service.

The more a tool can affect an individual's rights, opportunities or access to essential services, the more important it becomes to examine its purpose, limitations, oversight and accountability arrangements.

2. The accountability gap: can people find out what happened?

A report published on 20 October 2026 by the Public Law Project, funded by the Nuffield Foundation, examined transparency and accountability in public-body automated and automated-assisted decision-making.

The report, Missing Pieces: Transparency and Accountability in Public Body Automated (and Automated Assisted) Decision-making, argues that people can struggle to discover whether an automated tool was used in a decision affecting them. They may also find it difficult to understand how the tool operated or establish whether it treated them fairly.

This creates a practical problem. A person cannot effectively question a system they do not know was involved. They may not know what information was considered, whether an error occurred or which organisation is responsible for correcting it.

Transparency should therefore mean more than publishing a general statement that an organisation uses AI. People need understandable information about the tool's role in the relevant process and how they can raise concerns.

3. The UK's Algorithmic Transparency Recording Standard

The UK Government has developed the Algorithmic Transparency Recording Standard, commonly known as the ATRS. It provides a structured way for public-sector organisations to publish information about algorithmic tools, including what they do, why they are used, how they influence processes and what risks or safeguards have been identified.

The standard covers more than systems described specifically as AI. It can also apply to statistical models and other algorithmic tools that influence public decisions.

Under the published policy, ATRS recording is mandatory for all government departments and certain arm's-length bodies delivering public or frontline services or directly interacting with the public. Within those organisations, the requirement applies to tools that significantly influence decisions with public effect or directly interact with the general public, subject to the standard's scope and exemptions.

The ATRS is an important transparency mechanism, but it is not a guarantee that every use of AI across every public body is publicly recorded. The scope, applicable requirements and published records need to be considered when examining a particular service.

Nor does publishing a record automatically demonstrate that a tool is fair, accurate or appropriate. Transparency is a foundation for scrutiny; it does not replace testing, oversight, complaints processes or legal accountability.

Read the official Algorithmic Transparency Recording Standard guidance.

4. Who is responsible when AI gets it wrong?

Responsibility should be established before a tool is introduced, not worked out for the first time after someone has been harmed.

A public body may purchase a tool from a private supplier, use a system developed by another department or rely on an AI service integrated into existing software. These arrangements can involve several organisations, but they should not leave the person affected unsure where to turn.

Clear governance should identify:

• Who approved the tool and the purpose for which it is used.
• Who is responsible for the service or decision it supports.
• Who checks performance, accuracy and possible unfair outcomes.
• Who investigates errors, complaints and suspected harm.
• Who can pause or stop the system if serious problems arise.
• How the public can request an explanation or challenge a decision.

The Government's Ethics, Transparency and Accountability Framework for Automated Decision-Making emphasises testing, fair services, clear responsibility, safe data handling, understandable explanations, legal compliance and future-proofing.

Human oversight must be meaningful. A staff member who simply accepts a computer-generated recommendation without the time, information or authority to question it may not provide effective oversight.

5. Human review must be more than a formality

Public bodies may use AI to support professional judgement, but a human name attached to the final decision does not automatically make the process fair.

Meaningful review requires the decision-maker to understand the tool's role, consider the underlying evidence and recognise when the system may be wrong. They need the authority to disagree with the output and enough time to do so properly.

For significant decisions, organisations should be able to explain in plain English how the tool contributed to the process. They should also consider whether the system's recommendations are reliable for different groups of people and whether relevant information has been omitted or misunderstood.

Where a tool is unsuitable, an alternative process may be necessary. Where an error is discovered, the organisation needs a way to correct the record, reconsider the decision and address any resulting harm.

6. What rights do people have when decisions are automated?

Data protection law is an important part of the accountability picture. The Data (Use and Access) Act 2025 changed the UK's framework for certain decisions made solely through automated processing of personal information.

The changes allow solely automated decisions with legal or similarly significant effects in a wider range of circumstances, but safeguards remain important. These include providing information about significant decisions, enabling people to make representations, allowing human intervention and providing a way to contest a decision where the relevant rules apply.

The precise legal requirements depend on the decision, the information being processed, the applicable legal basis and whether additional protections apply to special-category data. Not every AI-assisted decision is a solely automated decision, and not every use of AI has a significant effect on an individual.

People should not assume that every decision involving AI is unlawful. Equally, organisations should not assume that describing a process as “AI-assisted” removes their responsibilities.

For anyone affected by a decision, practical steps may include asking whether an automated tool was used, requesting an explanation of its role, seeking human review and using the relevant complaints or appeal process.

Read the Government's explanation of the Data (Use and Access) Act 2025.

7. The impact on disabled and neurodivergent people

Public services should be accessible to people with different communication styles, disabilities, learning needs and levels of digital confidence. AI could help by making information easier to find, offering alternative formats or supporting staff with routine administrative work.

However, automated processes can create additional barriers if they assume everyone communicates or processes information in the same way.

A person might struggle with an inaccessible online form, an automated telephone system, a rigid identity-checking process or a tool that misinterprets their communication. A system trained on incomplete or unrepresentative data may also perform less reliably for some groups.

For disabled people, the ability to explain circumstances, provide information in an accessible format and ask for appropriate human support can be particularly important.

Public bodies should consider accessibility from the beginning, test tools with the people who may use them and provide suitable alternatives. They should also assess whether automated systems could create discriminatory effects and how those effects will be identified and addressed.

Technology should make services easier to access, not create another gate that people must get through before they can receive support.

8. Privacy, data quality and public trust

AI systems may process large amounts of personal information. If the data is inaccurate, out of date, incomplete or used in an inappropriate way, the resulting recommendations may be misleading.

Public bodies should understand what information a tool uses, why it is needed, how it is protected and whether it is suitable for the decision being supported.

They should also consider whether people can understand the use of their information and how to raise concerns. Where suppliers are involved, contracts and governance arrangements should make responsibilities clear and support appropriate scrutiny.

The Government's Data and AI Ethics Framework emphasises privacy, fairness, transparency and protection from harm. Its principles are relevant throughout a system's lifecycle, from defining the problem and selecting a tool to testing, deployment, monitoring and eventual retirement.

Trust should be earned through reliable services, clear explanations and evidence that problems are taken seriously. It should not depend on the public simply accepting that a computer must be right.

9. Following the money: what should public AI funding pay for?

Public investment in AI should not be assessed only by the number of systems purchased or the speed with which a service can process information.

Responsible implementation also requires money for testing, accessibility, staff training, data protection, monitoring and independent evaluation. Organisations need the capacity to investigate mistakes and change or stop systems that do not work as intended.

Funding decisions should consider:

• Pre-deployment testing: evidence that the tool is suitable for its intended purpose.
• Accessibility: testing with disabled people and others who may face barriers.
• Human oversight: sufficient staff, training and authority to review important outputs.
• Transparency: public information about significant uses of algorithmic tools, in line with applicable requirements.
• Redress: accessible complaints, review and appeal arrangements.
• Ongoing monitoring: checks for errors, changing performance, security problems and unintended effects.
• Independent evaluation: evidence about whether the investment improves outcomes and what problems remain.

These safeguards are not merely extra costs attached to innovation. They are part of the work required to operate consequential public services responsibly.

10. A practical checklist for public bodies

Before introducing or expanding an AI or algorithmic tool, a public body should ask:

• What specific public-service problem is the tool intended to solve?
• Is AI appropriate, or could a simpler process achieve the same result?
• What decisions will the tool influence, and how serious could the consequences be?
• Who is accountable for the system and the decisions it supports?
• Has the tool been tested for accuracy, security, accessibility and potential bias?
• Can staff explain its role and question its recommendations?
• Will the use of the tool be disclosed in the appropriate public records?
• Can affected people ask for an explanation and challenge a decision?
• Are privacy and data protection requirements being met?
• How will the organisation monitor performance, investigate complaints and correct errors?
• Will the public be able to see evidence of benefits as well as risks?

Conclusion: public responsibility cannot be automated away

AI may help public services process information and support staff, but the consequences of public decisions remain real for the people affected by them.

Transparency helps people understand how technology is being used. Human oversight helps ensure that recommendations are properly considered. Data protection, accessibility and fair processes help protect people's rights. Clear accountability makes it possible to identify who must act when something goes wrong.

These protections must be built into procurement, funding, implementation and ongoing monitoring. They cannot be added as an afterthought once a system is already influencing people's lives.

For AM-MediaSpace, responsible AI means treating technology as an assistant rather than an unquestioned authority. That principle applies not only to creative work but also to the public systems that shape people's opportunities, security and access to support.

The funding test: When public money pays for AI, is enough being invested in transparency, accessibility, human oversight and the ability to challenge mistakes — and can the public see the evidence?

Sources and further reading

• Public Law Project — Missing Pieces: Transparency and Accountability in Public Body Automated and Automated-Assisted Decision-Making (20 October 2026)
• UK Government — Algorithmic Transparency Recording Standard: Guidance for Public Sector Bodies
• UK Government — Ethics, Transparency and Accountability Framework for Automated Decision-Making
• UK Government — Data and AI Ethics Framework
• UK Government — Data (Use and Access) Act 2025: Data Protection and Privacy Changes
• Information Commissioner's Office — Rights Related to Automated Decision-Making, Including Profiling
• UK Parliament — Written Answer on Automated Decision-Making Systems in the Ministry of Justice (17 July 2026)

Editorial note: This article provides general information, not legal advice. The applicable rules depend on the specific system, decision and organisation. Public-sector transparency requirements and data protection safeguards have defined scopes and should be checked against current official guidance.