AI Unauthorized Practice

State of play.

• AI-assisted pro se litigation has moved from pleadings to trial. A Georgia state court litigant proceeded to jury trial after using AI to prepare his case — the first documented instance of AI-assisted self-representation reaching the trial stage, signaling that judicial management of generative AI in pro se matters now extends beyond written filings into live courtroom work .
• AI-assisted pro se filings are a measurable, accelerating volume problem. Federal pro se plaintiff filings rose from 11.33% pre-AI to 16.94% post-AI, with AI-generated complaints climbing from near-zero in 2019 to over 18% of all pro se complaints by 2026; Employment and Fair Housing Act cases show the sharpest increases — up 49% and 69% respectively — and federal judges have flagged the surge as an "existential threat" to docket management .
• Pennsylvania has applied existing professional-licensing statutes to an AI chatbot. The Pennsylvania Department of State and State Board of Medicine filed a Commonwealth Court action against Character Technologies, alleging its "Emilie" character violated the Medical Practice Act by presenting as a licensed psychiatrist and providing a false license number — among the first state enforcement actions targeting an AI companion platform for unauthorized practice of medicine .
• ABA Formal Opinion 512 and its state adoptions remain the ethics baseline but are contested. Mississippi Ethics Opinion No. 267, adopted verbatim from ABA Formal Opinion 512, permits reduced verification for familiar legal AI tools — a standard critics argue is contradicted by the Stanford hallucination data the opinion itself cites (17-33% hallucination rates across leading legal AI platforms) .
• For counsel advising law firms, lenders, or AI platform developers, the practical baseline is that AI-assisted pro se filings are now standard operational friction requiring early-scrutiny protocols, while the unauthorized-practice perimeter is actively expanding through state enforcement — and ethics opinions set a permissive floor, not a safe harbor.

Where things stand.

• ABA Formal Opinion 512 is the governing ethics baseline, but its reduced-verification permission is under documented pressure. The opinion establishes duties of competence, confidentiality, output verification, reasonable billing, and informed consent; its permission to reduce verification for familiar tools is the live fault line, with a Stanford study cited within the opinion finding 17-33% hallucination rates in leading legal AI systems .
• State bar adoptions are fragmenting the verification standard. Mississippi adopted ABA Formal Opinion 512 verbatim as Ethics Opinion No. 267; whether other states adopt, modify, or reject the reduced-verification permission will determine whether a national standard emerges or the landscape splinters .
• Consumer AI platforms create privilege-waiver exposure for both attorneys and clients. In United States v. Heppner (S.D.N.Y.), a federal court held that AI-generated documents created using Claude were not privileged because the tool was not a lawyer and was not used at counsel's direction; practitioner advisories now treat client-side AI misuse — uploading privileged documents to ChatGPT or Claude — as a distinct, operationalized waiver risk .
• ABA Model Rule 1.6(c) is the operative confidentiality hook. Privacy toggles and similar safeguards in consumer AI platforms do not satisfy the ethical standard for preventing unintended disclosure; inputting client data into public AI systems without enterprise-grade data agreements is a Rule 1.6 exposure .
• AI impersonation of licensed professionals is a regulated and actively enforced category. Pennsylvania's Commonwealth Court action against Character.AI is the most concrete enforcement action; New York has enacted statutes restricting AI impersonation of licensed professionals; the FTC has pursued injunctions against "robot lawyers" .
• New York has a pending legislative framework that would create chatbot-proprietor liability for AI providing legal or medical advice. The proposed statute targets the structural question the Pennsylvania enforcement action raises: whether AI platforms bear civil liability — not just regulatory exposure — when their systems hold out as licensed professionals .
• Sanctions for AI-generated hallucinations are documented at the highest levels. Chief Justice Roberts cited AI-generated fake citation incidents in his 2023 Annual Report; federal courts continue to sanction attorneys; courts are now mobilizing staff to recognize AI-generated materials in pro se filings .
• Judicial AI use is outpacing the ethical framework governing it. Approximately 60% of judges report using AI tools while attorneys continue to face sanctions for AI-assisted work product errors — a structural asymmetry with no current resolution .

Latest developments.

• Georgia pro se litigant proceeds to jury trial after using AI to prepare his case — first documented instance of AI-assisted self-representation reaching the trial stage; courts have not yet resolved what disclosure obligations or accuracy standards apply to AI use in live proceedings rather than written filings .
• AI-assisted pro se borrower litigation is creating measurable enforcement friction for lenders and banks, with AI-generated complaints rising to over 18% of all federal pro se complaints by 2026 and federal judges flagging the volume as an existential docket-management problem; early scrutiny and expedited motion practice are now recommended operational responses .
• Pennsylvania sues Character.AI in Commonwealth Court, alleging the "Emilie" character violated the Medical Practice Act by presenting as a licensed psychiatrist with a false license number — among the first state enforcement actions applying existing professional-licensing statutes to an AI companion platform .

Active questions and open splits.

• What disclosure obligations apply to pro se AI use in live proceedings? Courts have addressed AI in written filings through sanctions and disclosure requirements, but the Georgia trial case exposes a gap: no settled standard governs AI use during examination, argument, or other real-time courtroom work .
• Can existing professional-licensing statutes reach AI systems that hold themselves out as licensed professionals? Pennsylvania's Commonwealth Court action against Character.AI tests whether the Medical Practice Act applies to an AI character presenting as a psychiatrist — a threshold question that will determine how broadly state licensing boards can deploy existing law against AI platforms .
• Does experience with a legal AI tool justify reduced verification? ABA Formal Opinion 512 and Mississippi Ethics Opinion No. 267 say yes — but the Stanford hallucination data they cite says the premise is unsound. No court has yet tested whether reliance on this guidance defeats a malpractice or sanctions claim .
• What is the scope of privilege waiver when a client uploads attorney work product to a consumer AI platform? Heppner establishes the waiver theory for AI-generated documents not created at counsel's direction, but the outer boundary — how much client-side AI use taints the privilege — is unsettled .
• Where does AI legal assistance end and unauthorized practice begin? The FTC's "robot lawyer" injunctions, state impersonation statutes, and the New York chatbot-liability bill define the regulatory perimeter from one direction; the boundary for AI tools that provide legal information short of representation remains contested across jurisdictions .
• What ethical framework governs judicial AI use? With approximately 60% of judges using AI tools and no parallel sanctions regime, the asymmetry between judicial and attorney AI use is unaddressed — and whether judicial AI use in decision-making triggers disclosure or recusal obligations is an open question .
• Will the ABA revise Formal Opinion 512 in response to documented criticism? The reduced-verification permission is under pressure from both the hallucination data and the accelerating sanctions record; whether the ABA issues supplemental guidance or courts simply override it through sanctions is the near-term resolution path .

What to watch.

• How Georgia courts manage the AI-assisted pro se trial — specifically whether they impose disclosure requirements, accuracy standards, or sanctions for AI-generated errors in live proceedings, which would set a bellwether for trial-stage AI management nationally .
• Commonwealth Court's treatment of Pennsylvania's Character.AI action — whether the court holds that existing medical-licensing statutes reach AI systems, which would open the door to analogous unauthorized-practice-of-law enforcement against AI legal platforms .
• Whether New York enacts the chatbot-proprietor liability statute, which would create a civil damages framework layered on top of existing regulatory enforcement .
• Additional state bar ethics opinions adopting, modifying, or rejecting the ABA's reduced-verification permission — Mississippi is the first but unlikely to be the last.
• Further judicial decisions applying or extending Heppner's privilege-waiver theory to client-side AI use in civil matters.
• Whether courts or disciplinary bodies address the judicial AI use asymmetry through disclosure requirements or recusal standards.