AI can draft an audit workpaper, a brief, a claim decision, or a client letter. It cannot sign one. In every licensed profession, the person who signs stays responsible for the work, whether a person or a model wrote the first draft. Regulators, bar authorities, and courts have said so in plain words since 2023. So as AI does more of the work, the scarce product becomes a review worth signing. Firms that design that review well can use far more AI, safely.
The Rule Is the Same in Every Profession
Audit. Only licensed CPAs may issue a report on financial statements. The Uniform Accountancy Act calls that rule "the cornerstone prohibition" of the act. PCAOB staff guidance says a team member who uses a generative AI tool "is still responsible for the results and documentation" of the work. Reviewers should apply the same diligence as for work done without AI. The PCAOB's amendments on technology-assisted analysis apply to audits of fiscal years beginning on or after December 15, 2025. In October 2026, the AICPA issued a practice aid. Its chief auditor said the auditor "remains responsible for the audit and the audit opinion."
Tax. The IRS Office of Professional Responsibility issued AI guidelines in June 2026. Due diligence "requires verifying the accuracy of facts, citations, and calculations produced by AI." Practitioners should document their AI use and their verification.
Law. The ABA's Formal Opinion 512 (July 2024) applies the existing rules on competence, confidentiality, communication, supervision, and fees to generative AI. The State Bar of California updated its practical guidance in May 2026 to cover agentic AI. It says a lawyer "remains fully responsible" for any output and work product made with AI. The more autonomy an AI system has, the more supervisory controls and verification the lawyer needs.
Insurance. The NAIC model bulletin asks insurers for an AI program run by senior management accountable to the board. It also makes insurers answer for vendors: decisions made with a vendor's AI must meet the same legal standards as the insurer's own. As of October 2026, 29 jurisdictions had adopted it, and California, Colorado, New York, and Texas had their own insurance AI rules.
Securities. FINRA's Notice 24-09 reminds firms that the supervision rule applies to generative AI. Communication rules apply whether a person or a tool wrote the text. FINRA's 2026 oversight report suggests keeping prompt and output logs and having people review model outputs.
Healthcare. For Medicare Advantage, CMS says AI "alone cannot be used as the basis to deny admission" to a hospital. An adverse medical-necessity decision must be reviewed by a physician or other appropriate professional. California's SB 1120, in effect since January 2025, requires the same for health plans in that state.
Courts Enforce It
In Mata v. Avianca (June 2023), a federal court sanctioned lawyers who filed a brief with citations invented by ChatGPT. The judge wrote that using a reliable AI tool is not improper, but existing rules "impose a gatekeeping role on attorneys."
Cases like it keep growing. The AI Hallucination Cases Database kept by researcher Damien Charlotin listed 2,241 cases worldwide as of October 29, 2026. Lawyers filed 886 of them, and self-represented parties most of the rest.
Some courts now ask for a certificate. A 2026 standing order in the Eastern District of Texas requires a certificate from any filer who used AI. The filer must state that they "independently verified the accuracy" of what the tool drafted.
Why the Signature Becomes the Scarce Product
When AI drafts most of the work, the cost of the first draft falls close to zero. The cost of the signature does not. A signer must know enough to catch what the AI got wrong, and carries the liability if they miss it.
The essay The Commanding Heights by Will Manidis and Joe Roberts expects licensing to keep its value for this reason. In their view, the license becomes a surety and an audit signature. That has a practical consequence for firms. The bottleneck is no longer how much work juniors can draft. It is how much work qualified signers can review well. Our post on human-in-the-loop agent design covers why review quality falls when it is not designed.
A Review Worth Signing: The Checklist
Each item below comes from one or more of the sources above.
| # | Practice | Where it comes from |
|---|---|---|
| 1 | Set review depth by the cost of an error. More autonomy and higher stakes need more control | NAIC bulletin; California Bar guidance |
| 2 | Verify every citation, number, and fact before it leaves the firm | Mata v. Avianca; IRS OPR guidelines |
| 3 | Decide the individual case. Do not let a general model decide for a specific client or patient | CMS Medicare Advantage FAQ |
| 4 | A qualified person signs adverse and attest outcomes | Uniform Accountancy Act; CMS rule; California SB 1120 |
| 5 | The reviewer can override, and overrides are logged | EU AI Act Articles 14 and 26 |
| 6 | Same diligence for AI-assisted work as for any other work | PCAOB staff guidance |
| 7 | Keep the evidence trail: prompts, outputs, model version, reviewer | FINRA 2026 report; IRS OPR guidelines |
| 8 | Record the judgment, not only the approval. Write what was checked and why it passed | PCAOB staff guidance; California Bar guidance |
| 9 | Diligence and contract your AI vendors, with audit rights | NAIC bulletin; FINRA Notice 24-09 |
| 10 | Protect confidential data before it goes into any tool | ABA Opinion 512; IRS OPR guidelines |
| 11 | Disclose and certify where a court or regulator requires it | E.D. Texas standing order |
| 12 | Bill for real time, and make no inflated claims about AI | California Bar guidance; IRS OPR guidelines |
| 13 | Name an owner at management or board level, with policy and training | NAIC bulletin; FINRA supervision rule |
Design the Work Around the Signer
Three design choices make a signature both faster and safer.
- Machines check before people sign. Automatic citation checks, tie-outs, and rule checks run first, so the signer spends time on judgment, not on typos. See our guide to AI agent audit trails.
- The evidence travels with the draft. The signer sees the sources, the calculation, and what the AI changed, in one place.
- The review is recorded. A short note of what was checked becomes the documentation regulators ask for, and the training data for better checks next quarter.
Build Sign-Off Into Your AI Work With OpenNash
OpenNash builds AI workflows for CPA firms, law firms, insurers, and advisers with the review step designed in. Our forward-deployed engineers:
- Map each workflow to the rule that governs its sign-off.
- Add automatic checks for citations, numbers, and policy rules before a person reviews.
- Keep the evidence trail: every prompt, output, model version, and reviewer decision, in an audit log you own.
- Route each case by risk, so signers spend their time where an error costs the most.
- Train your reviewers and signers on what AI gets wrong in your work.
Book a call with an OpenNash forward-deployed engineer or email us at hello@opennash.com. In the first session, we review your top AI use cases and the sign-off rules that apply. Then we show you which ones deliver the highest business value and the fastest ROI.