AI Notetaker for Accountants Tax: The Audio Path Decides

10/15/2026

By Kyle Nelson, Founder, Fazit

ogTitle: "AI Notetaker for Accountants Tax: § 7216 Has an Opinion"

An AI notetaker for accountants doing tax work is not a productivity decision first. It is a confidentiality decision with a criminal statute attached. IRC § 7216 makes unauthorized disclosure of tax return information a misdemeanor, and § 6713 adds a civil penalty per disclosure. The notetaker that survives that test is the one that never creates an audio file and never sends the conversation to a vendor.

What § 7216 Actually Says

For a tax return preparer, the rule is not a vague professional standard. IRC § 7216 makes unauthorized disclosure of tax return information a misdemeanor. IRC § 6713 adds a civil penalty for each unauthorized disclosure.

A cloud AI notetaker sitting on a client call is a disclosure event. The audio, or a transcript of it, leaves the preparer's machine and lands on a vendor's infrastructure. That vendor may then send it to a transcription service, a model provider, or a subprocessor. Each hop is a disclosure the preparer authorized by installing the tool. The accountant-specific post on this site walks through the statute in detail. The short version: the posture that avoids the question is one where the audio never leaves the Mac and never becomes a file.

This is not legal advice. Consent and professional-conduct duties apply to the person running the call, and a tax preparer's obligations under § 7216 and Circular 230 sit on top of any tool choice.

The Tools Tax Firms Are Told to Use

Intuit's Tax Pro Center has published a piece on practical AI uses for tax professionals. It names Fathom and Otter.ai as tools that can capture discussions, identify action items, and help build client meeting summaries. The framing is productivity: less typing, fewer missed commitments, faster follow-up.

Canopy goes further. Its Notetaker automatically joins, records, transcribes, and summarizes client and internal meetings. Canopy says its accounting-aware summaries flag client intent, tax risks, and commitments relevant to tax and advisory work. It links meeting data directly to a client record in its practice-management software.

Vinyl, listed in Xero's AI app collection for accountants, records, transcribes, and files meeting notes while syncing client records from XPM.

The note is useful. The recording is the problem.

The Recording Is the Artifact You Cannot Control

A tax client call contains more than the return. It contains cash-flow anxiety, a divorce, a business partner dispute, a payroll mistake, a question about whether something should be reported. That conversation is tax return information in the broadest sense, and it is exactly what § 7216 protects.

When a notetaker records the call, the preparer has created a second record. That record sits on a vendor's servers under a retention policy the preparer may not have reviewed. Vinyl's own guidance to accounting firms says to verify whether recordings, transcripts, and summaries are used to train AI models and where the data is stored or processed. That is the right question, and it is a question you should not have to ask about your own client call.

The Otter.ai wiretap litigation shows what happens when the recording becomes the center of a legal dispute. The complaint alleges a bot recorded a sales call without everyone's consent and fed the audio into model training. No court has ruled yet. But the case is a checklist of what a cloud notetaker does with your calls, and a tax preparer reading it should recognize the exposure.

Consent Does Not Fix the Architecture

Some vendors advise firms to inform participants and obtain consent before recording. That is necessary. It is not sufficient.

Consent makes the recording lawful in many jurisdictions. It does not make the recording disappear. The audio file still exists on a vendor's infrastructure. The transcript still exists in a database. The summary still exists in a practice-management system. Each of those is a record the preparer may have to produce, explain, or defend.

The Granola lawsuit makes the point from the other direction. A proposed class action filed in July 2026 argues that a bot-free notetaker left other participants with no way to know a recording was running. The complaint does not establish liability. It does establish that the consent question applies to every notetaker, including the ones that do not show a bot in the participant list.

For a tax preparer, the cleanest posture is one artifact: a note the preparer retains under their own policy, with no recording behind it. That is the architecture Fazit uses.

What Fazit Does Differently

Fazit is a native macOS menubar app. It captures the call app's audio and the microphone directly through a Core Audio process tap. No bot joins the call. No other participant sees anything.

The audio exists only in a fixed-size RAM ring buffer with no write API. It is destroyed on every exit path, including errors. Every note records audio_retained: false in its frontmatter. Transcription runs on-device using Parakeet via CoreML, and note generation runs against a local model on localhost. The output is a plain Markdown file written into an Obsidian vault folder. There is no Fazit database holding a copy.

The architecture post explains why this is an invariant, not a marketing claim. The audio never becomes a file, so there is nothing to store, nothing to delete later, and nothing to subpoena.

The Comparison That Matters

Tool	Bot joins call	Audio leaves device	Recording survives call	Note location
Canopy Notetaker	Yes	Yes	Yes	Canopy practice management
Vinyl	Not stated	Yes	Yes	Vinyl + XPM sync
Fathom	No	Yes	Yes	Fathom cloud
Otter.ai	Yes	Yes	Yes	Otter cloud
Fazit	No	No	No	Your Obsidian vault

The first four tools all write a useful note. The difference is the artifact they create along the way. Fazit is the row where the recording was never eligible to exist.

The IRS Already Has Enough Data

The Texas Society of CPAs reported in January 2026 that the IRS uses AI and data analytics to select tax returns for audit and detect errors and fraud. Almost all tax returns are submitted electronically or scanned into IRS computers, creating years of tax records available for analysis.

That is the data the IRS already has. The preparer's job is not to add to the pile. A cloud notetaker that stores client-call audio and transcripts on a vendor's servers creates a second, less protected copy of the same sensitive information. The preparer who avoids that copy has one less thing to explain in an audit, a data breach, or a client dispute.

The Checklist for Tax Firms

Before putting any AI notetaker on a client call, a tax preparer should be able to answer four questions:

Does a bot join the call as a participant?
Does the audio leave the preparer's machine?
Does a recording survive after the note is written?
Where does the note file live, and who holds a copy?

If the answer to any of the first three is yes, the tool has created a disclosure surface that § 7216 cares about. If the answer to the fourth is a vendor's database, the preparer has outsourced custody of their own work product.

Fazit answers: no bot, no audio leaves the Mac, no recording survives, and the note is a Markdown file in the preparer's own Obsidian vault. That is the architecture a tax preparer can defend.