Getting the best from AI – safely and responsibly

AI tools can boost productivity, creativity and efficiency, but they also bring risks around data security, accuracy and compliance.

Here, our Group AI Delivery Lead, Liz Millington, left, explains how to use tools such as Copilot and Claude responsibly – from protecting sensitive information and checking AI-generated outputs carefully to ensuring every use case is approved, accountable and aligned with our policies.

AI tools like Copilot and Claude are now part of day-to-day working life for many of us. Used well, they save time and help us think through problems differently. Used carelessly, they can put our data, our clients and our practices at risk.

The most important rule: watch what you put in.

Anything you type into a free-to-use AI tool can potentially be seen by the company running it and used to train their models. That includes tools you might already use at home or have downloaded yourself. Just because a tool is free and easy to access does not mean it is safe for work use.

That means you should never paste in:

• Personal details about clients, patients or colleagues

• Practice or company names

• Financial information such as pricing or costs

• Passwords or IT system details

• Patient record details that could identify the animal or their owner. General symptoms or conditions are fine, as long as nothing could be traced back to a specific patient or client.

  
  

A good test before you type anything in: would you be comfortable if that information appeared on a public website? If not, leave it out.

It is also worth thinking about combinations of information. A job title plus a location plus a team name might not seem sensitive on its own, but together it can identify a specific person. Think about what you are sharing as a whole, not just line by line.

If your use case genuinely needs sensitive data, stop and contact the AI Delivery Team first and we can help you find a safe route.

Always treat AI output as a first draft

AI tools can sound confident while being completely wrong. This is called hallucination. Before you share or publish anything AI has helped produce, proofread it, fact-check it and make sure the tone fits VetPartners. Where AI is helping you reach a decision, the final call sits with a person, not the model.

Get into the habit of asking the tool where it got its information from, not just when something looks wrong. AI can produce output that sounds completely plausible but is based on an incorrect or invented source. A quick prompt asking for the source can save real problems down the line.

A few other things to keep in mind....

• AI can carry bias from the data it was trained on. Apply your own judgment.

• Generated text and images can raise copyright issues. Check before using.

• If you use AI to generate an image, label it clearly as AI-created. Do not publish AI-generated images without making that visible to the reader.

• Any new AI use case, including paid tiers of any tool, needs approval from the AI Steerco before you sign up.

  
  

Governance matters - and it is there to protect all of us

VetPartners has Group GenAI guidance in place for a reason. AI is moving fast and the risks are real. The guidance is not there to slow you down, it is there to make sure we get the most out of these tools without taking risks we cannot see.

Before you start using a new AI tool at work, ask yourself:

• Has this tool been approved for use at VetPartners?

• Am I on a paid or enterprise plan, or a free version?

• Do I know where my data goes when I use it?

  
  

If you cannot answer those questions, check before you start. Free versions of tools almost always mean your inputs can be used to train the model. Paid or enterprise versions often have stronger data protections, but they still need AI Steerco approval before you sign up.

We are here to help

If you are unsure about anything, get in touch at ai@vetpartnersgroup.com. We will help you find the right approach without slowing you down.