Your Team Is Already Using AI — Have You Set the Boundaries?

An employee uses AI to improve an email. A manager asks it to draft a policy. A care coordinator tries it to summarise information before a meeting. Each task may seem straightforward, but the decisions behind it matter: which system is being used, what information is entered and who checks the result?

If you have not discussed AI with your team, you may not know where it is already becoming part of their working day.

Recent coverage on The Daily Round has highlighted concerns about risks from autonomous AI and warnings about AI-generated health advice on social media. For care leaders, these debates raise an immediate management question: do employees understand the boundaries for using AI in your organisation?

Clear expectations can help staff use useful tools confidently while protecting the people, information and decisions entrusted to them.

Start by finding out what people are using

Ask your team which AI tools they use, what they use them for and where they find them helpful. Include AI features built into existing software, as well as separate chatbots and applications.

Make the purpose of the conversation clear. You need an accurate picture of working practices so that you can assess them and provide guidance. Employees may be less willing to explain what they do if the discussion begins with an assumption that every use is misconduct.

Ask staff to demonstrate tasks using fictional examples. This can help managers understand both the benefits and the points at which information leaves an approved system or an output influences a decision.

Define the tasks you will allow

Give employees recognisable examples of acceptable use. Drafting a general meeting agenda or improving the wording of a non-confidential email involves different considerations from summarising someone’s care records or suggesting how to respond to a change in their health.

Decide which activities are permitted, which require approval and which must remain outside the tool’s use. Be specific about the systems covered: approval to use one product for one purpose should not be treated as permission to use any AI application.

Some specialist systems may be assessed and approved for sensitive work. That approval needs to relate to the particular task, information and safeguards involved.

Be explicit about what staff can enter

Employees need clear instructions about personal, confidential and commercially sensitive information. That includes information about people receiving care, relatives, employees and the organisation itself.

Removing a name does not necessarily make a document anonymous. Details about someone’s circumstances, location or health may still make them identifiable.

A practical starting rule is that staff should not enter confidential or identifiable information into an unapproved AI tool. Where a system is approved for handling particular information, explain the conditions of that approval and who employees should ask when they are unsure.

Managers should understand how the proposed service handles information, including access, storage and any use of inputs for training. Those questions belong in the organisation’s assessment of the tool.

Treat a polished draft as something to check

AI can produce fluent wording that contains errors, assumptions or invented details. A professional tone does not establish that the content is accurate.

An AI-assisted email should be checked against the original facts and the intended message. A policy draft needs review against the organisation’s actual procedures, responsibilities and relevant requirements. A reference or source should be opened and verified before anyone relies on it.

Explain what checking means for each task. Asking someone to “review the output” is too vague if they do not know which facts, commitments or instructions require particular attention.

Set a higher threshold for care documentation

Care records influence what happens to a person. Any approved use of AI in preparing or summarising them therefore needs a clearly defined process and appropriate professional oversight.

Staff should never accept wording that adds an observation they did not make, changes the meaning of an account or presents an assumption as something that happened. A summary also needs checking for information it has left out.

Be clear about whether your organisation permits AI assistance with care documentation at all, which system may be used and who must verify the final record. Employees should understand how to correct an error and escalate a concern.

Keep responsibility clear

Allocate responsibility for checking and approving AI-assisted work. An employee should know whether they can send a draft themselves, need a manager’s review or must refer it to someone with specialist expertise.

The reviewer needs enough knowledge to assess the content. Passing an AI-generated policy to another person who cannot judge its accuracy adds little reassurance.

The Daily Round’s coverage of Sir Chris Whitty’s call for gradual training in clinical risk included his warning that public-facing AI can present conclusions with misleading certainty. That is a useful reminder for leaders to preserve room for uncertainty, questions and professional judgement when introducing technology.

Show staff what good use looks like

Training should use examples employees recognise. Demonstrate how to improve a general email without including confidential details, how to spot unsupported claims in a policy draft and how an apparently helpful summary can alter meaning.

Give staff a named person to contact when they are uncertain. Explain what to do if they enter information into the wrong system, discover an inaccurate output or find that an approved tool behaves unexpectedly.

People need to be able to report mistakes promptly so that the organisation can respond.

Review the boundaries as use develops

Assign someone responsibility for maintaining your AI guidance and reviewing approved tools. Revisit it when systems, features or intended uses change, and ask employees whether the instructions remain clear in practice.

A useful policy should help someone decide what to do during an ordinary working day. Can they use this tool for this task? Can they enter this information? Who must check the result? Where should they raise a concern?

If your team can answer those questions, you have given them a practical foundation for responsible AI use.

Stay informed with the stories shaping health and social care.

Register with The Daily Round for the latest sector news, inspections & reports, wellbeing content, workforce insights, expert features and practical updates—delivered free to your inbox.