MNPI and AI Tools: Staying Compliant When AI Touches Material Non-Public Information

A simple guide for compliance officers. Keep AI inside your information barrier and out of trouble with MNPI rules.

MNPI and AI tools are a risky mix if you get it wrong. MNPI is information that is both important and not public. AI is great at the exact tasks that touch it.

Think of summarizing a data room or comparing two indentures. That overlap is the danger. The wrong tool can move MNPI outside your walls in one click.

The good news is simple. The rules already cover this, and the fix is clear. This guide shows you how to use MNPI and AI tools together, safely.

What counts as MNPI in an AI context

MNPI is information a reasonable investor would find important. It is also not yet public. In an AI context, the test is the same as anywhere else.

What changes is speed. A single prompt can carry MNPI into a system that stores or trains on it. That is the risk to control.

  • Unannounced M&A, financings, and restructurings
  • Draft or pre-release financial results
  • Draft credit ratings and rating-committee materials
  • Term sheets, offering documents, and indentures before announcement
  • Any confidential client data a counterparty would find material

The MNPI rules that apply to AI tools

The rules never mention AI. They do not need to. They are technology-neutral, so they cover AI prompts like any other channel.

FINRA said this plainly in Regulatory Notice 24-09 in June 2024. The notice adds no new rules. It confirms the old ones reach generative AI.

  • Rule 10b-5 and the misappropriation theory: no trading on or misusing MNPI.
  • Regulation FD: no selective disclosure of MNPI.
  • FINRA Rule 3110: supervise for MNPI misuse and keep information barriers.
  • FINRA Rule 5280: do not trade ahead of research reports.

Map which AI workflows can touch MNPI

You cannot control what you have not mapped. So list every place AI could see MNPI. This is more than the obvious chatbot.

Look at AI features inside other tools, too. An AI note-taker on a deal call counts. So does an AI feature in your CRM or a coding assistant that reads internal files.

For each tool, ask three questions. Could it ever see MNPI? Where does the data go and who can read it? Does the tool retain or train on inputs?

  • List every AI tool and every AI feature inside other software.
  • Flag the ones that could plausibly receive MNPI.
  • Document data flow, retention, training behavior, and access for each.
  • Send MNPI workflows only to approved, no-training, access-controlled AI.
The riskiest AI tools are often the ones nobody "deployed" — the auto-joining note-taker, the browser extension, the personal account. Find these first.

Keep MNPI-touching AI inside the information barrier

An information barrier is only as strong as its weakest channel. An AI tool can be that weak point. Its logs or training pipeline may leak across the wall.

The fix is to control three things. Control the endpoint. Control retention. Control who can read the logs.

Use a simple rule. If a person could not see the deal, they should not see the prompts about the deal.

  • Use no-training, access-controlled AI for any MNPI workflow.
  • Limit AI workspaces and logs to walled staff only.
  • Turn off AI features that broadcast outputs firm-wide.
  • Prefer zero data retention or self-hosting for the most sensitive deals.

Recordkeeping and supervision of AI

AI does not escape books-and-records rules. If AI is used for firm business, the records may need to be kept. This can fall under Rule 17a-4, Rule 204-2, or Rule 17g-2.

That points to one answer: log it. You must be able to retain and review AI use where the rules require it. You must also supervise it for misuse, like any other channel.

Build logging in from the start. Adding it after an exam request is much harder.

  • Decide which AI interactions are records under 17a-4, 204-2, or 17g-2.
  • Capture prompts and outputs for those workflows in a reviewable form.
  • Add AI to your supervision for MNPI misuse and insider trading.
  • Keep the logs access-controlled so they do not breach the wall.

Common MNPI and AI mistakes to avoid

Most incidents are not high-tech. An analyst pastes a draft into a personal chatbot to fix the wording. An AI assistant quietly records a deal call.

The pattern is always the same. It is convenience plus an uncontrolled tool.

The defense is just as simple. Give people a fast, safe tool. Then block the risky ones.

  • Personal-account use — block consumer AI and offer a sanctioned tool.
  • Auto-joining note-takers — control which assistants join sensitive calls.
  • Outputs in shared folders — limit where AI output is stored.
  • Browser extensions and plug-ins — govern these like any AI tool.

Conclusion: use MNPI and AI tools the safe way

MNPI and AI tools can work together when you set clear limits. Map where AI meets MNPI. Keep that AI inside your wall. Log and supervise it.

The biggest win is also the easiest. Block consumer AI and give your team a safe alternative. That removes most of the risk in one move.

Do this and your analysts get AI's speed without breaking the rules. Use the related guides below to set up the right tools.

Frequently Asked Questions

  • No. Using AI on MNPI is not illegal by itself. The risk is misusing MNPI or breaking your information barrier. For example, sending it to a tool that trains on inputs, or whose logs are visible outside the wall. A private, no-training, access-controlled AI tool is usually fine for MNPI work.
  • Not a standalone rule. FINRA Regulatory Notice 24-09 from June 2024 reminds firms that existing rules apply to AI. These include supervision (3110), trading ahead of research (5280), communications, and recordkeeping. The notice adds no new rules but confirms the old ones reach AI.
  • Often, yes. If the AI interaction relates to firm business, it may be a record. That can fall under Rule 17a-4 for broker-dealers, Rule 204-2 for advisers, or Rule 17g-2 for NRSROs. Build logging into MNPI-touching AI so you can retain and review what the rules require.
  • Yes. An AI note-taker on a deal call records MNPI. It then stores that data wherever the tool stores data. That may be outside your wall and in a system that retains or trains on inputs. Control which assistants can join sensitive meetings and where their output goes.
  • Block consumer AI tools and give staff a safe, no-training alternative. Most MNPI and AI incidents come from convenience. Someone uses a personal account because it is fast. Remove that reason and you remove most of the risk.
  • An AI tool that sees MNPI must sit inside the barrier. That means a no-training endpoint, an access-controlled workspace, and logs limited to walled staff. If someone who could not see the deal can see the prompts about it, the barrier is breached.

Let your team use AI on sensitive work, safely

Layer3 Labs sets up no-training, access-controlled AI that stays inside your information barrier. We add the logging and supervision your MNPI rules require.

Book a free MNPI-and-AI review