A ChatGPT policy (or broader AI policy) should set out clear guidelines and boundaries for the circumstances in which the use of AI is and isn’t permitted. The context of an employee’s specific role will be crucial in shaping these guidelines.

In crafting your ChatGPT policy, the starting point will be to identify:

• Tasks or roles where employees might be able to leverage AI to enhance productivity and efficiency, provided appropriate quality measures are in place; and
• Those where AI assistance might not be appropriate, such as activities involving sensitive data handling, decision-making processes with significant consequences, or tasks requiring human creativity, intuition or critical thinking. For these activities, you may wish to prohibit the use of AI altogether, or impose strict guidelines.

A complaints handler, for instance, might effectively utilise AI to draft initial responses to customers, resulting in significant time savings and increased productivity. Your policy may permit this, though it may be wise to state that employees are expected to thoroughly review and validate any AI-generated responses for accuracy and appropriateness before sending them to ensure the final response aligns with the organisation’s standards and maintains a human touch.

A greater degree of caution might be warranted when it comes to content creators, as AI-generated material could inadvertently include copyrighted works or factual inaccuracies, posing legal and reputational risks for both the employee and the employer. You may wish to allow the use of AI for certain content creation tasks only, or introduce extra vetting procedures such as fact-checking and proofreading to minimise potential problems.

For high-stakes roles, where the potential consequences could be severe, a judicious and meticulous approach should be taken when considering the use of AI tools. Legal advisors, for example, would need to exercise extreme caution when relying on ChatGPT for generating advice to clients or preparing court submissions, as highlighted by a recent case in which a lawyer submitted court documents which included AI-generated, fictitious case law. The lawyer had assumed the AI’s output to be entirely factual and had failed to recognise its potential for errors.

Indeed, it is important to keep in mind that ChatGPT relies on the data it has been trained on, which may not always be up to date or entirely accurate. As such, it is essential for employers to explicitly communicate in their policies that AI is fallible and should not be assumed to be true or reliable. To this end, it is a good idea to reinforce your policy through training to help employees understand the limitations of AI and exercise critical judgment when utilising AI-generated content.

By carefully defining these parameters through policies and other communications, and reinforcing them in practice, employers can promote a culture of responsible and effective AI use while mitigating potential risks and liabilities.

As well as defining where AI should and should not be used, your policy should also:

• Consider data protection and the security of personal and confidential company information.
• Set out your expectations regarding originality and intellectual honesty.
• Be clear about the consequences of not complying with your company’s rules on AI use.

If employees violate the policy by using ChatGPT in authorised ways or fail to properly vet AI-generated work, then disciplinary action could follow, in the usual manner.