Does your AI policy help or block?

Most AI policies were written for a chat window. An agent that works in your files, runs code and leaves a history is a different thing, and many policies have not met it yet. It is just early.

Eight questions place your organisation's rules on a ladder, from no rules at all to rules that say yes, carefully. Answer for the rules as they are, not as you would like them. This is not legal or compliance advice: whoever owns your policy has the last word.

1. Q1Is there a written AI policy where you work?

   Not that I have found, or just "don't", with no reason given.There is a short note, mostly about not pasting things into chatbots.Yes, a written policy I could find and quote.Yes, with a named owner, and it is reviewed on a schedule.
2. Q2Does it treat a chatbot and an agent differently?

   I am not sure it covers agents at all.As far as I can tell, all AI is one chat window.It mentions agents or coding assistants, without much detail.Yes. Agents have their own rules: which folders, what they may run.
3. Q3What data could you use with an AI tool today?

   I don't know, so I use none.Public data, I think. Nobody has said otherwise.It is written down which kinds of data may go where.The tiers are written down, and there are synthetic stand-ins for the sensitive work.
4. Q4Are approved tools named?

   No, and I am not sure who I would ask.There is a list of banned tools, not approved ones.One tool is approved: a chat assistant.Several are named, including one that works in files, with the settings to use.
5. Q5You want to try something the policy doesn't cover. What happens?

   I would not know where to start, so I leave it.I email someone and wait.There is a named contact for exceptions.There is a pilot route: scoped, time-limited, on public or synthetic data.
6. Q6If an agent did some work, could someone review it afterwards?

   Only from what I remember.From the chat history, if I kept it.From version control: I commit the agent's changes.It is required: commits, logs, and a record of what the agent was asked.
7. Q7Who signs off outputs that AI helped produce?

   Nobody in particular. They go out like anything else.I do, informally.A named reviewer, as for any other analysis.A named reviewer, told which parts AI helped with, as the policy requires.
8. Q8Does your policy say what good use looks like?

   If it exists, it is mostly don'ts.A few don'ts, then "use your judgement".It gives some examples of acceptable use.It has worked examples, and says who to ask about the grey areas.

The button opens an email to us, copied to Rabbie, our robot, who keeps the diary. Your result travels in it so we can prepare. You can change or delete it before you send, and nothing leaves your browser until you do. Whatever you choose from the ways to work with us, the first call asks what you're allowed to use, and the work is designed within that. What an agent does that a chatbot doesn't is in Chatbots talk. Do-bots work.

Back to all the ways to work with us →