PART TWO OF THREE
Closed-group drafts are travelling further than they should. I think I know why.
In financial services, you've got industry bodies and regulators running closed-group conversations all the time. A regulator and an industry association will circulate draft documentation for comment to named individuals at each of a handful of banks. That's part and parcel of how legislation gets shaped before anyone sees it.
I come from the banking environment. If you were part of that closed group, you knew. You didn't leak it and you didn't ask around outside it. Nobody had to spell that out.
Something has shifted in the last two years. Restricted drafts are being shared well outside the groups they were written for, and the frequency of it is going up.
“There's no respect anymore for confidentiality, or people don't understand why certain documents cannot be shared with a greater group.”
People have become blasé about information because of the way they now work. A confidential draft goes into ChatGPT to be summarised. It goes into the most accessible or familiar AI tool to be explained. Nothing happens. No alarm, nobody objects. Once that feels normal, passing it to somebody outside the group feels normal too.
These are senior professionals who know the rules perfectly well. No covering note, no hesitation, nothing to suggest anything unusual has happened. Three years ago, not one of them would have done it. The status of those documents hasn't changed but the habit around them has.
Confidential in red bold letters across the top used to carry the whole instruction, because everybody in the room shared the same assumptions about what it meant. It doesn't carry it anymore.
“It's no longer enough to say confidential in red bold letters across the top.”
If I'm the regulator circulating that draft, I need to spell it out. Not to be shared externally, which it probably already says. Then the part that's missing: whether it may go into a large language model at all, and if so, which ones. Some institutions have their own internal models and can reasonably permit those. Others should say plainly that this document doesn't go into any external tool. It looks laborious. It's a lot less laborious than what happens otherwise.
“Everybody's so hyped up about having to use AI. We're not having express conversations about when not to use it.”
The assumption underneath all of it is that people already know where the lines are. They do not know.
On the flip side you've got media houses that will publish anything for clickbait, and a lot of these documents are controversial. Take something like expropriation without compensation here in South Africa. A very early draft, something people still need to collaborate on and contribute to, lands on a news feed and the purpose of the document is gone, because it was never ready for public consumption. The people working on it in good faith go quiet. The drafters get more cautious.
“There's a reason these documents can't be leaked before they're ready for public comment. It leads to longer lead times for important pieces of legislation to see the light of day.”
Now add fabricated documents to that same environment. Early leaks, good fakes and official releases all circulating together, and working out which is which becomes a specialist job. That's where we're heading, and it's why verification is going to be worth more than speed.
Read Part One here.
--
This article was written by Beulah Lüttig, Director | Head of Operations at Afriwise Law Explorer (Pty) Ltd
.png)
