Your employees already use AI. The question is whether this happens with or without agreements.
A workable AI policy specifies what data should never go into an AI tool, what tools are allowed, when a human checks the result, and that AI output always remains the responsibility of the person using it. One A4 is enough; a ban doesn't work, because then it happens unnoticed.
STEP BY STEP
This is how you approach it
- Assume it's already happeningStudy after study shows that employees use AI without reporting it. A policy that pretends this is not the case is immediately outdated.
- Determine which data should never be releasedCustomer data, personal data, source code, contracts, medical or financial information. This is the most important rule and the only one that should be absolute.
- Designate allowed toolsBusiness accounts where input is not used for training. This varies per provider and subscription type: check instead of assuming.
- Determine who remains responsibleThe employee using AI output remains responsible for the content. “The AI said so” is not an accountability.
- Determine where a human should checkEverything that goes outside, everything with legal or financial consequences, and all the figures. You always check facts from a language model.
- Tell everyone and keep it shortNo one reads a ten-page policy. One A4 with five lines is remembered and therefore adhered to.
PITFALLS
Why banning doesn't work
A ban moves usage to private accounts on private devices. Then you have the same risk, without visibility and without agreements.
Moreover, you miss out on functionality that your competitor uses. Steering is more effective than holding back, and it is the only option that gives you vision.
- Don't ban — then it happens unnoticed.
- Make data boundaries absolute — the only hard rule.
- Name permitted tools — check the training policy.
- Do invest responsibility — at the user.
- Keep it short — one A4 is read.
- Do check facts — always, without exception.
FREQUENTLY ASKED QUESTIONS
More about AI policy
Is an AI policy mandatory?
Rules are coming that will impose requirements on the use of AI systems, depending on the application. Apart from that, it is simply sensible: the biggest risk is an employee who pastes customer data into a free tool. Have the legal side tested for your situation.
Can employees use ChatGPT?
You decide that. A workable line: yes for rewriting your own texts and looking up general information, no for everything with customer data in it, and always check before it goes out.
Do you write such a policy?
We think along about the technical side and can say which tools apply which training policy. A lawyer is the right party for the legal formulation.
CONTINUE READING
Of course, continue reading about this subject.
AI with appointments?
We think along about which tools are safe and where the boundaries should be.