What is an AI tax copilot? A practical guide for tax professionals

The short answer: a copilot assists, the professional commands

An AI tax copilot is an assistant that works alongside a tax professional across the entire practice—not a single feature bolted onto one screen, and not a system that runs the firm on its own. Where a narrow tool automates one task, a copilot is present throughout the work: it helps assemble a return, surfaces the context you need about a client, accelerates research, and drafts the emails and memos that fill a firm's day. The defining characteristic is the relationship, not the technology. You decide; the copilot assists. You approve; the copilot proposes.

The name is borrowed deliberately from aviation, and the analogy is worth taking seriously. A copilot reduces workload, flags what needs attention, and hands the controls to the person in command—it does not fly the plane alone or overrule the pilot. In a tax practice, the person in command is a credentialed professional who reviews, applies judgment, and signs. That structure is not a marketing preference; it is what the rules governing tax practice require, and it is what keeps a copilot useful rather than dangerous.

The IRS Office of Professional Responsibility has said this plainly. In mid-2026 it issued guidance on the use of artificial intelligence in federal tax practice, and as the Journal of Accountancy summarized, the core message is that "AI should augment—not replace—professional judgment," that final decisions must rest with qualified professionals, and that practitioners bear full responsibility for any AI-assisted work product. That sentence is the entire operating manual for a copilot. This guide unpacks what it means in the day-to-day work of a firm.

What an AI tax copilot actually is—and is not

It helps to separate three things people casually lump together: a feature, a copilot, and an autopilot.

A feature versus a copilot

A feature does one thing—extracts figures from a W-2, or checks a return against last year. A copilot is broader: it spans intake, preparation support, client management, research, and communication, and it carries context from one task to the next so you are not re-explaining the client to a new tool at every step. The value of a copilot is coherence. It knows this is a return you are mid-way through reviewing, that the client has an outstanding document request, and that last week you flagged a K-1 question—so its suggestions are situated in the actual work rather than generic.

A copilot versus an autopilot

This is the distinction that matters legally and ethically. An autopilot would make decisions and act without a human in the loop—determining a filing position, sending a return, answering a substantive client question—on its own authority. A copilot never claims that authority. It proposes a draft, a summary, a next step, or a flag, and waits for a professional to accept, edit, or reject it. Everything a copilot produces is provisional until a person approves it. A vendor that markets an "autopilot" for tax work is describing something the profession's own standards do not permit, because responsibility for the work cannot be delegated to software.

What a copilot is not

A copilot is not a preparer of record, not a signer, and not a source of tax authority. It does not hold a PTIN, it cannot appear before the IRS, and its output is not itself a citation you can rely on. When a copilot tells you the treatment of an item, that statement is a lead to be checked, not an answer to be trusted. Keeping this clear in your own mind is the single most important habit for using one well.

What a copilot does across the practice

The reason "copilot" fits better than "tool" is breadth. A well-built copilot shows up in four areas of the practice, and in each one it does the assembling and drafting while leaving the judgment to you.

Preparation support

On the preparation side, a copilot classifies incoming documents, extracts the figures, and helps populate a draft return in your tax software, then flags what looks missing or inconsistent so gaps surface as questions rather than silent errors. This is preparation support—the copilot gets the return organized and pre-checked so you open it ready to review, not ready to type. The mechanical share of preparation is exactly the work that automation should absorb; the interpretation and sign-off remain yours. If you want the detail on that boundary, our companion guide on whether AI can prepare tax returns walks through it end to end.

Client context

Second, a copilot assembles the context you would otherwise dig for: prior-year returns, open items, the last conversation, documents received and outstanding, deadlines, and what changed year over year. Instead of opening five systems to reconstruct where a client stands, you get a situated summary. This is where the "360" in a practice-wide view earns its name—the copilot connects the record so preparation, research, notices, and planning draw on the same client picture rather than fragments. We cover that connected-context model in depth in how connected client context ties research, notices, and planning together.

Research acceleration

Third, a copilot accelerates research by finding relevant authority, summarizing it, and pointing you to the sections that matter—turning a starting question into a shortlist of sources far faster than a cold search. This is genuinely useful and genuinely dangerous, and the danger is the reason for the strict verification rule in the next section. A copilot can fabricate a citation or misread a holding with complete confidence. Used as a way to find primary authority you then read yourself, research acceleration is a real advantage; used as a substitute for reading the authority, it is a liability.

Drafting

Fourth, a copilot drafts the written work a firm produces constantly: the client email requesting a missing 1099, the engagement follow-up, the plain-language explanation of an IRS notice, the internal memo summarizing a research question. Drafting is where copilots save the most everyday time, because so much of a firm's communication is necessary but not novel. The professional edits, corrects, and approves before anything leaves the office—but starting from a solid draft instead of a blank page is a meaningful lift across a season.

The line a copilot must not cross

Everything above is assistance. The line begins where responsibility begins, and it is drawn by the rules that govern practice before the IRS—not by product design choices.

Circular 230 duties do not change because you used a tool

Treasury Department Circular 230 establishes standards of competency, diligence, and ethical conduct for practitioners, and the Office of Professional Responsibility enforces them. When the OPR turned to AI in 2026, it did not create a new regime—it emphasized that long-standing duties such as due diligence, competence, and confidentiality remain unchanged when a practitioner uses AI. The copilot is inside your practice, but the obligations sit on you. A copilot cannot exercise due diligence for you; it can only help you do it.

The profession's own standard: reliance never absolves you

The AICPA reached the same conclusion in its Statements on Standards for Tax Services. The revised standards effective January 1, 2024 added Section 1.4, Reliance on Tools, which directly addresses software, research aids, and AI. Its principle is stated cleanly: a member may reasonably rely on tools used in providing tax services, but—as the standard puts it—use of the tool does not absolve the member of their professional obligations. Practitioner commentary on the standard is blunt that review is the final necessary step for all reliance on tools: a tool should enhance the professional's understanding, not supplant their judgment. That is the copilot boundary in a single sentence.

Where the line falls in practice

Concretely, a copilot must never be the one to choose a filing position, decide a matter of professional judgment, answer a substantive tax question to a client as if it were authority, or send a return or filing on its own. Those are command decisions. The copilot can prepare the ground for every one of them—assemble the facts, draft the language, surface the options—but a credentialed professional makes the call and owns it. When you evaluate a product, look for a visible boundary that shows what the AI did and hands you the decision, not one that quietly acts on your behalf.

Verify against primary sources, not the copilot

If there is one rule to carry out of this article, it is this: you verify a copilot's output against primary sources, never against the copilot itself. A copilot's confidence is not evidence. Its summary of a code section is a pointer to the code section, not a replacement for reading it.

This is exactly where the IRS OPR guidance is most pointed. Its warning, as widely reported, is that blind reliance on what AI yields—especially when the underlying logic or sources are unclear—may constitute unreasonable reliance, and that practitioners must verify AI-generated content by checking citations, reviewing the cited authorities, and confirming calculations and inputs. The OPR stated that human scrutiny and editing are essential to ensure correctness. Large language models can and do produce fabricated citations and confident errors; a figure or a case that "looks right" is not the same as one you have traced to its source.

In workflow terms, the discipline looks like this:

1. Treat every copilot output as a draft. A research summary, a drafted email, a proposed entry—each is provisional until you have checked it.
2. Follow every citation to its source. If the copilot cites a code section, revenue ruling, or publication, open the actual authority on IRS.gov or your research service and confirm it says what the copilot claims. Verify the citation exists before you verify what it holds.
3. Reconcile every figure to a document. A number the copilot placed on a return gets checked against the source W-2, 1099, or statement—the same way you would check a data-entry clerk's work.
4. Judge the conclusion yourself. Whether an item belongs on a Schedule C or as other income, whether a position is supportable, whether a credit's due-diligence requirements are met—these are yours to decide, informed by but not delegated to the copilot.

None of this is unique to AI. It is the same supervisory discipline a firm already applies to a preparer's or an intern's work. Circular 230 itself contemplates reasonable reliance on the work product of another person when the professional uses reasonable care in supervising and evaluating it. A copilot is another such source of work product—capable, fast, and in need of exactly that supervision.

Data governance: §7216, confidentiality, and your WISP

A copilot touches client tax data constantly, which means adopting one is a data-governance decision before it is a productivity decision. Three obligations frame it.

IRC §7216: use and disclosure of return information

Internal Revenue Code §7216 imposes criminal penalties on preparers who knowingly or recklessly disclose or use a taxpayer's return information for purposes other than preparing that return, unless an exception or the taxpayer's consent applies. The Treasury regulations were written for the software era and govern the use and disclosure of return information in electronic and software-based preparation. Applied to a copilot, the practical questions are direct: does the tool use client data only to serve that client's engagement, or does it also train external models, share data with third parties, or retain it for other purposes? If a copilot's data use extends beyond preparing the client's return, you may need specific, informed §7216 consent. This is a procurement question you settle before routing a single client file through the tool.

Confidentiality and enterprise-grade tools

The IRS OPR guidance reinforced the same point from the confidentiality side, warning practitioners to handle client data using only secure, enterprise-approved AI and to avoid public or unsecured platforms that expose confidential information. Pasting a client's figures into a consumer chatbot is precisely the misuse the guidance targets. A copilot built for tax practice should keep client data inside a controlled environment with defined access, encryption, and retention—not send it to a general-purpose service whose data practices you cannot see.

The FTC Safeguards Rule and your WISP

Paid tax preparers are treated as "financial institutions" under the Gramm-Leach-Bliley Act, which places them under the FTC Safeguards Rule. Every firm must maintain a Written Information Security Plan (WISP); the IRS and Security Summit remind practitioners that this is a federal mandate with no small-firm exception, and provide a template in Publication 5708 and guidance in Publication 4557, Safeguarding Taxpayer Data. Any copilot you adopt becomes part of the environment your WISP must account for. That means its encryption, access controls, service-provider terms, and data retention belong in the plan—and in the vendor evaluation—not as afterthoughts. Our security checklist for AI software turns this into a concrete list of questions to ask.

How to evaluate an AI tax copilot

Because a copilot spans the practice, evaluate it on more than accuracy on any one task. The table below maps common copilot capabilities against what the tool does and what the professional owns—the boundary a good copilot makes explicit.

Capability	What the copilot does	What the professional owns
Preparation support	Classifies documents, extracts figures, drafts a return in your tax software, flags missing and inconsistent items	Reviews every material figure against source documents; interprets, corrects, and approves the return; signs it
Client context	Assembles prior returns, open items, deadlines, and year-over-year changes into one situated view	Judges what the context means for the engagement and what to do next
Research acceleration	Finds candidate authority, summarizes it, and points to relevant sections	Reads the primary source, confirms the citation exists and holds what is claimed, and decides the position
Drafting	Produces first-draft emails, notice explanations, and memos in plain language	Edits for accuracy and tone, confirms every factual claim, and approves before it is sent
Data handling	Keeps client data in a controlled, encrypted environment with defined access and retention	Confirms §7216 and WISP compliance, sets consent where required, and vets the vendor's data use

Questions that separate a copilot from a gimmick

Beyond the table, four questions cut to whether a copilot fits a real firm. First, does it carry context across tasks, or is each feature a silo you must re-brief? Second, is the human-approval step built in and visible—can you see what the AI did and why before you accept it? Third, does it slot into the tax software and workflow your firm already uses, so a drafted return lands in Drake, ProSeries, or Lacerte rather than a parallel system that adds re-entry? Fourth, can you trace every figure and citation back to its source for verification? A copilot that answers these well is an assistant your professionals can supervise; one that cannot is a black box you would be signing behind.

A day in the practice with a copilot

Consider how a copilot changes an ordinary morning at a small firm. The following is illustrative, not a statistical claim.

A preparer opens the day's queue. The copilot has already classified and extracted the documents a client uploaded overnight, drafted the return in the firm's tax software, and flagged two items: a 1099-DIV referenced on the brokerage summary that has not arrived, and dividend income well above last year. Alongside the return, a client-context panel shows the prior-year figures, the open document request, and a note from last week's call. The preparer reviews the flagged items, reconciles the entered figures against the source documents, and accepts the copilot's drafted email requesting the missing 1099—after editing a line for tone.

Mid-morning, a different client's return raises a question about whether a settlement payment is taxable. The preparer asks the copilot, which returns a summary and points to a code section and a revenue ruling. The preparer opens both on IRS.gov, reads them, confirms the copilot's summary is accurate on one point and incomplete on another, and reaches a conclusion the copilot did not fully supply. The research took fifteen minutes instead of an hour of cold searching—but the professional, not the copilot, decided the position and can defend it.

By afternoon, an IRS notice comes in for a third client. The copilot classifies the notice, drafts a plain-language explanation for the client and a first pass at a response, and pulls the relevant history into view. The professional verifies the notice type, corrects the draft, and approves the client communication—work that connects naturally to notice response and, when the season settles, to year-round planning conversations the copilot's assembled context makes easy to start.

Across the morning, the repetitive assembly and drafting compressed from hours to minutes, and the professional's time shifted to verification, judgment, and advice—the work only they can do and only they can sign for. That is the honest promise of an AI tax copilot: not a replacement for the professional, but an assistant that clears the friction so credentialed people spend their scarce hours where they are irreplaceable. Bringing that assistance together across preparation, context, research, notices, and planning is what Practice 360 is built to do.

Frequently asked questions

What is an AI tax copilot?

It is an AI assistant that works alongside a tax professional across the whole practice—supporting preparation, surfacing client context, accelerating research, and drafting communications—while the professional decides and approves everything. It proposes; you command. It is not an autopilot: it does not exercise judgment, sign returns, or take responsibility for the work.

Can an AI tax copilot make decisions on its own?

No, and it should not be designed to. Choosing a filing position, deciding a matter of professional judgment, and sending a return or filing are command decisions that must rest with a credentialed professional. The IRS Office of Professional Responsibility's 2026 guidance is explicit that AI should augment, not replace, professional judgment and that practitioners bear full responsibility for AI-assisted work.

Do I still have to verify what an AI copilot produces?

Yes—every time, against primary sources. You verify a copilot's output against the actual authority or source document, never against the copilot itself. Follow each citation to the code section, ruling, or publication on IRS.gov and confirm it exists and says what the copilot claims. Blind reliance on unverified AI output can constitute unreasonable reliance.

Is it safe to put client tax data into an AI copilot?

Only with a tool built for it. IRC §7216 restricts how you use and disclose return information, and the FTC Safeguards Rule requires a Written Information Security Plan covering any system that touches client data. Use secure, enterprise-approved tools—not public chatbots—and confirm the vendor's data use, retention, and whether §7216 consent is required before routing client files through it.

How is a copilot different from regular tax automation?

A single automation does one task; a copilot spans the practice and carries context from one task to the next, so preparation, client management, research, and drafting draw on the same client picture. The common thread is that both assist rather than replace—the professional reviews, judges, and signs in either case.

Sources and methodology

This article is based on published IRS guidance, the Internal Revenue Code preparer provisions, FTC and AICPA standards, and reporting on the IRS Office of Professional Responsibility's 2026 AI guidance. Any figures or scenarios are illustrative and labeled as such—they are not statistical claims. Rules are current as of publication and should be verified for the applicable tax year.

• IRS — Office of Professional Responsibility and Circular 230
• Journal of Accountancy — IRS Outlines AI Risks, Circular 230 Duties for Tax Practitioners
• AICPA — Statements on Standards for Tax Services (SSTS)
• The Tax Adviser — Understanding New SSTS Section 1.4, Reliance on Tools
• IRS — Section 7216 Information Center
• FTC — Safeguards Rule: What Your Business Needs to Know
• IRS — Security Summit: Tax Pros Must Have a Written Information Security Plan
• IRS — Publication 4557, Safeguarding Taxpayer Data (PDF)