Gross Misconduct for Using AI? Where Is the Proportionality?

A Derbyshire detective is being investigated over allegations involving the use of Microsoft Copilot to assist with policing decisions and processes, raising important questions about artificial intelligence, data protection and whether the response to its use is necessary and proportionate.

According to the Independent Office for Police Conduct (IOPC), Derbyshire Constabulary had authorised Copilot for limited administrative purposes, including drafting reports and summarising information. However, the officer is alleged to have used AI for risk assessments, policy decisions and justifying decisions, as well as entering restricted information.

These are serious allegations that deserve proper consideration, but they also highlight a much wider issue: policing needs to understand both the opportunities and dangers of AI before officers find themselves facing potentially career-ending proceedings.

AI Cannot Be Relied Upon to Make Policing Decisions

Artificial intelligence has enormous potential to reduce bureaucracy, speed up administrative processes and allow officers to spend more time doing the work they joined the police to do.

However, there is a significant difference between asking Microsoft Copilot to help draft a document and relying on it to determine whether a particular policing decision is lawful, necessary or proportionate.

General-purpose AI tools are not inherently designed or validated to make complex decisions involving UK legislation, police powers, safeguarding responsibilities or force policies. They can produce convincing responses that are inaccurate, outdated or based on information that is simply not applicable to the circumstances.

An officer asking AI whether an arrest is necessary, for example, might receive a detailed explanation that appears legally sound but overlooks crucial facts or misinterprets the relevant statutory criteria.

The danger is that AI can sound authoritative even when it is completely wrong.

Developing reliable AI-assisted policing tools requires considerably more than connecting a chatbot to the internet or uploading legislation. Individual processes need to be carefully designed, supported by authoritative information, tested against realistic scenarios and regularly reviewed as legislation and policies change.

Even then, AI should support an officer's professional judgement rather than replace it, with responsibility for operational decisions remaining firmly with the individual making them.

What Happens When Personal Information Is Entered Into AI?

Perhaps the greatest immediate concern is the potential disclosure of sensitive policing information.

Officers routinely handle personal details, intelligence, witness statements, criminal allegations and information concerning vulnerable individuals, much of which must be protected by strict confidentiality and data protection requirements.

When information is entered into an AI service, it may be transmitted to external servers and, depending on the provider, product and configuration, could be retained in logs, accessed by authorised administrators or subprocessors, or processed in other jurisdictions.

Some consumer AI services may use submitted information to improve their models, whereas enterprise products such as Microsoft 365 Copilot provide different contractual protections, including restrictions on using organisational prompts and responses to train foundation models.

However, those protections do not automatically make every use appropriate or compliant with policing requirements, particularly where information is confidential, operationally sensitive or entered into an unauthorised service.

There is also a risk that apparently anonymous information could identify someone when combined with dates, locations or circumstances, meaning that simply removing a person's name may not provide adequate protection.

Officers should never enter personal, identifiable or sensitive operational information into an AI tool unless their organisation has expressly authorised that specific use and established appropriate safeguards.

But Is Gross Misconduct Always the Answer?

We fully recognise that police officers must be accountable for their actions and that any inappropriate disclosure of confidential information deserves proper investigation.

However, the existence of an alleged breach does not remove the need to consider its seriousness, the circumstances surrounding it and the proportionality of the response.

In the Derbyshire case, the full evidence has not been publicly established, and it would be wrong to prejudge either the officer's conduct or the eventual outcome.

Nevertheless, we should be asking whether officers have received sufficient training, whether force policies are genuinely clear and whether less serious mistakes are being distinguished from deliberate or reckless wrongdoing.

A gross misconduct investigation can have profound consequences for an officer's career, reputation, family life and psychological wellbeing, particularly when proceedings become prolonged.

Where serious misconduct is reasonably suspected, formal investigation may be entirely justified, but the response should reflect the actual conduct and evidence rather than simply the fact that artificial intelligence was involved.

If policing expects officers to embrace innovation and work more efficiently, forces must provide approved technology, clear instructions and appropriate training rather than leaving individuals to navigate unfamiliar risks.

Pocket Sergeant Wants to Help Develop Safer AI Tools

At Pocket Sergeant, we believe there is an opportunity to develop AI-assisted tools that help officers navigate legislation, understand procedures and work through structured decision-making processes more efficiently.

However, these tools cannot be developed responsibly by software engineers alone, because each policing process requires specialist operational knowledge, carefully designed safeguards and rigorous testing.

We would therefore welcome conversations with experienced officers who have specialist knowledge of areas such as investigations, safeguarding, custody, traffic legislation and operational decision-making, and who would be interested in helping shape future tools.

It is imperative that no personal, identifiable or sensitive operational information is entered into experimental AI agents. Any development and testing would need to use fictional or appropriately anonymised scenarios, with suitable security and governance arrangements.

Our objective would never be to replace police officers' judgement, but to give them better access to accurate information and structured guidance when making decisions for which they remain accountable.

Policing Needs Innovation and Proportionality

Artificial intelligence is not going away, and policing cannot afford to ignore technology that could significantly reduce workloads and improve efficiency.

Equally, the risks surrounding inaccurate advice, confidentiality and data protection cannot be dismissed simply because the technology is convenient.

What concerns us is the possibility of officers facing the most serious disciplinary processes when better guidance, training or supervision might have prevented the problem in the first place.

Investigate genuine wrongdoing, absolutely, but ensure the response is proportionate and never forget the human being at the centre of the investigation.

Policing needs to embrace technology responsibly while protecting both the public and the officers expected to use it.