Book a principal
Exec x AISay No to Shadow AI

Day 4Material non public information entered into a consumer AI tool

AI generated studio photograph of a woman in her late thirties with dark hair tied back, wearing a grey collarless blouse, facing the camera directly against a plain pale grey background with a level, unreadable expression. A translucent panel across the lower half carries the caption in dark serif type: Our financial results were available three weeks before we published them. Below it, in small capitals: Stacy, Finance Director.

Our financial results were available three weeks before we published them

Stacy, Finance Director

Why this one is dangerous

Unpublished results are price sensitive. Once they are entered into an external service, the company can no longer say who holds them, where they are stored, or for how long.

Three weeks is the part that matters. That is the period in which the information was outside the company's control while remaining unpublished, and the period in which any dealing on it would have been dealing on inside information.

Stacy did nothing malicious. She used a tool that produced a better summary than the one she could write at eleven at night. No rule told her she could not.

How to combat it

  • Classify results, forecasts, board papers and draft announcements as restricted, and block those classes at the egress point rather than in a policy document.
  • Extend closed period controls to AI tools on the same footing as trading and communications. If the tool is not covered by the closed period, the closed period has a gap in it.
  • Add AI tool access to the insider list process, so the population that can reach unpublished numbers is a list you can produce.
  • Provide finance teams with a licensed instance carrying contractual no training terms and a defined retention period, before the reporting cycle rather than after it.
  • Brief the finance function specifically. Generic AI training does not cover the moment at eleven at night when the summary is due.