AI Meeting Notes for Expert Network Calls: How Hedge Funds Should Handle MNPI-Adjacent Conversations

Published September 19, 2026

September 19, 2026 · 10 min read

Expert network calls are one of the highest-value information channels a hedge fund analyst has — and one of the highest-risk. They sit exactly one bad question away from material nonpublic information (MNPI), and SEC exam staff already scrutinize them for tipping and trading correlations. When you drop a cloud AI notetaker into that workflow, you are not adding a productivity tool. You are creating a permanent, searchable copy of your firm's most sensitive conversations on a vendor's servers — a copy your Chief Compliance Officer (CCO) will have to supervise, retain, and potentially produce.

This guide is for the analyst, IT lead, or compliance officer at a hedge fund, long/short shop, or private equity firm who is evaluating an AI notetaker for use on expert calls, management team meetings, and diligence sessions. It walks through the actual regulatory surface, the architecture questions that matter, and why a botless, on-device approach reshapes the MNPI conversation.

Why Expert Network Calls Are a Regulatory Hotspot

The SEC's Office of Compliance Inspections and Examinations has repeatedly flagged expert networks in exam priorities. Compliance guidance from ACA Global on MNPI compliance is explicit: firms using expert networks should implement best practices such as logging and tracking calls, reviewing detailed call notes, and reviewing trading activity in publicly traded companies in industries similar to those discussed during expert network calls.

That review requirement is the exact reason firms want AI notetakers on these calls in the first place — durable, searchable notes make the compliance workflow tractable. But the same feature that makes AI transcripts useful for supervision makes them dangerous when they live on a third-party server outside your control.

The Skadden Warning: AI Tools and MNPI Are Now One Compliance Problem

In July 2026, Skadden published a client alert on when AI models access nonpublic information. The alert reframes the compliance question: broker-dealers and investment advisers must maintain policies reasonably designed to prevent misuse of MNPI, and firms risk scrutiny if AI tools foreseeably could use restricted data improperly, even absent actual trades.

The specific to-do list Skadden lays out — inventory data restrictions, segregate and permission AI access to nonpublic information, use explainability and audit trails, and ask governance questions before granting AI tools access — reads like a checklist that a cloud AI notetaker fails by design. A vendor that ingests every meeting an analyst joins and stores it indefinitely on shared infrastructure is not "segregated" or "permissioned." It is the opposite.

Rule 204-2 and the Recordkeeping Trap

Investment advisers have to think about AI transcripts through the lens of the Books and Records Rule. Smarsh's Rule 204-2 overview lays out the obligations: create and retain specific categories of records, maintain them for at least five years with the first two years in an easily accessible location, and preserve electronic communications such as emails, texts, social media posts, and chats that relate to advisory services.

Where AI notetakers get thorny is the "sent or received" question. Steel Eye's recordkeeping analysis for SEC-registered advisers summarizes the practical distinction: if content sits inside an application and is not actively sent, it may not qualify as a covered communication under Rule 204-2, but the moment an adviser emails an AI-generated summary or posts a transcript excerpt in a chat, it becomes a written communication subject to regulatory requirements. Steel Eye cites the conservative Stark & Stark view that transcripts of advisory discussions should be retained even if not shared externally, because they might be the sole detailed record of the advice given.

Skadden's earlier September 2024 analysis makes the same point about AI functionality in Zoom and Microsoft applications and their intersection with broker-dealer and investment adviser communications rules, noting that the SEC has been amending recordkeeping rules to keep pace.

The Otter Litigation Changed the Threat Model

Any compliance officer evaluating AI notetakers in 2026 needs to have the In re: Otter.ai Privacy Litigation case on their radar. Hintze Law's September 2026 analysis notes that on August 13, 2026 the U.S. District Court for the Northern District of California allowed significant portions of the proposed class action to proceed, with plaintiffs alleging that the company's meeting assistant joined virtual meetings, recorded and transcribed conversations in real time, collected voice-related information, retained meeting content, and used that information to improve its products and machine-learning systems.

Jackson Lewis's analysis in the National Law Review frames the lesson simply: organizations leveraging these technologies must balance efficiency with compliance, ensuring that recording, consent, and data-use practices align with evolving privacy and other laws. For a hedge fund, that translates: the vendor holding your expert call transcripts is not just an operational vendor. It is a litigation target and a subpoena surface.

The Bot-vs-Botless Question Is Actually a Consent Question

Beyond the data-flow question, there is a workflow question that hits expert calls especially hard. MeetingNotes' 2026 asset-manager buyer guide spells it out: a recording bot that appears as a visible participant in a Zoom or Teams call alerts every attendee — including counterparties — that the call is being captured. For management team calls, LP conversations, and operational due diligence sessions, this creates relationship and consent complications. Botless tools capture audio natively from the desktop without joining as a participant.

On an expert network call, a visible bot is worse than an awkward feature — it can reshape the call itself. Experts who see "Fireflies Notetaker" in the participant list may hedge language, cut short answers, or ask their compliance team to sign off. You get worse information and a worse compliance record. The Dark Reading coverage of the tl;dv incident hammered the point: if an AI notetaker shows up in a call you did not invite, that is a red flag — the bot is a participant.

What the Buyer Checklist Actually Looks Like

Adapted from vendor evaluation guidance published by Fellow for private equity firms and hedge funds, an MNPI-aware buyer checklist should require: written confirmation in the DPA that meeting content is never used to train or improve the vendor's AI models, documented data storage location, an up-front sub-processor list, current SOC 2 Type II certification, a GDPR-ready DPA, and a security questionnaire the vendor can turn around in a reasonable window.

That is the floor. It is not the ceiling — none of those controls address the architectural question of whether a copy of your expert call transcript exists on a vendor's server at all.

Cloud vs On-Device: The MNPI Comparison That Matters

Control	Cloud AI notetaker (typical)	On-device AI notetaker
Processing location	Vendor cloud (US or EU region)	Analyst's Mac or iPhone
Transcript custodian	Vendor + your firm	Your firm only
Bot in participant list	Usually yes	No (botless native capture)
Training on your content	Depends on DPA — historically default-on	No model training pipeline
Subpoena surface	Firm + vendor	Firm only
Sub-processor list to audit	Full stack of cloud sub-processors	None for the transcript itself
Retention configurability	Vendor-defined tiers	Firm-defined at endpoint
Data residency questions	Recurring — depends on vendor region	Not applicable — data stays on device

The point of the table is not that on-device transcription magically produces "compliance." It does not. Compliance is your CCO's determination. The point is that on-device processing removes one large category of controls you would otherwise have to build, audit, and defend to an examiner.

How the Vendor Landscape Handles MNPI Today

The cloud vendor policies you would inherit if you dropped Otter, Fireflies, or Zoom AI Companion into an expert call are all public. Otter.ai's privacy policy and Fireflies' privacy policy should be reviewed clause-by-clause by counsel before either tool touches a call with an expert. Zoom's privacy statement is similarly relevant if you plan to use Zoom AI Companion for capture.

Regulators are also moving. Social Europe's analysis of AI note-takers at work flags that if meetings were recorded or transcribed without participants' knowledge, this may constitute a personal data breach under Articles 33 and 34 of the GDPR, triggering obligations to notify the supervisory authority and, in some cases, the data subjects themselves. For US hedge funds with EU counterparties on an expert call, that is a live exposure.

How Basil AI Solves This: On-Device Capture for MNPI-Adjacent Calls

Basil AI is an on-device AI meeting notetaker for iPhone and Mac. Audio capture, transcription, and summarization all run on the analyst's device using Apple's on-device Speech framework and the Apple Neural Engine — described in Apple's privacy overview. There is no vendor server holding a copy of the expert call. There is no cloud training pipeline that could ingest MNPI. There is no bot in the participant list.

For a hedge fund, that changes the compliance conversation in three concrete ways. First, the vendor sub-processor question collapses — there is no sub-processor for the transcript, because the transcript never leaves the device. Second, the subpoena surface for the artifact is your firm alone, which is what your GC wants when a call gets pulled into an SEC exam. Third, the workflow is botless: the expert never sees a recording participant, so the call itself is not distorted by capture.

None of that makes Basil AI "compliant" — compliance is always your CCO's determination applied to your firm's specific policies. What on-device architecture does is give you a smaller, cleaner surface to reason about.

For related reading on the buyer-side of this question, see our guides to the post-Otter procurement checklist for enterprise buyers, AI meeting notes for compliance officers in financial services, and AI meeting notes for asset managers and MNPI.

A Practical Rollout Plan for an Investment Team

Step 1: Inventory the calls

Before choosing a tool, map which analyst workflows the notetaker will touch. Expert network calls, management calls, sell-side calls, LP calls, and internal investment committee meetings each have a different MNPI profile. Your CCO should confirm which categories are in scope.

Step 2: Segregate by risk

Skadden's guidance to segregate and permission AI access to nonpublic information applies here. Not every meeting should feed the same tool with the same retention settings. Expert calls should have the tightest capture posture; internal team standups can be looser.

Step 3: Prefer botless and on-device where possible

For calls with counterparties, prioritize botless capture. For the highest-sensitivity subset, prefer on-device processing so the transcript never becomes a vendor artifact.

Step 4: Write retention into policy

Your Rule 204-2 program should say — in writing — how long transcripts and summaries are kept, where, and who can access them. Ad hoc per-user settings are the failure mode examiners look for.

Step 5: Train the analyst team

The best architecture fails if an analyst forwards an expert call transcript into a group chat. Training should cover when a transcript becomes a "sent or received" communication and how to handle MNPI hits mid-call.

Bottom Line

Expert network calls are exactly the kind of conversation where the architecture of your AI notetaker matters more than its marketing. Cloud tools give you a searchable transcript and a second custodian. On-device tools give you a searchable transcript and no second custodian. For an MNPI-adjacent workflow, that difference is the entire compliance conversation.