Bar officials stopped an AI from arguing in court

The conversation

In January 2023, Joshua Browder, chief executive of the startup DoNotPay, said his company would have an AI feed a defendant arguments through an earpiece during a real traffic-ticket hearing, scheduled for 22 February in a California courtroom. NPR reported the plan called for the defendant to wear smart glasses that would record the proceeding and generate responses using AI text tools, including ChatGPT. Browder told NPR that after word spread, “multiple state bars have threatened us,” including a warning that “a referral to the district attorney's office and prosecution and prison time would be possible,” and DoNotPay dropped the plan as a result. DoNotPay's own account is that the system was called off before any argument was made.

What the documents show

NPR's reporting documents statements from both sides. Browder is quoted saying the threat of prosecution, not a settled ruling, was enough: “Even if it wouldn't happen, the threat of criminal charges was enough to give it up.” The State Bar of California's Chief Trial Counsel, George Cardona, is quoted saying the organization “regularly let[s] potential violators know that they could face prosecution in civil or criminal court, which is entirely up to law enforcement” — a statement of general practice, not a finding that DoNotPay had violated the law. The State Bar's executive director, Leah Wilson, separately describes a broader trend of “well-funded, unregulated providers rushing into the market for low-cost legal representation.”

The system boundary

The unrealized system was designed to place an AI-generated argument inside a proceeding that otherwise runs on licensed human representation, with no described role for a lawyer once the earpiece began feeding responses. State bar organizations exist to police that boundary: they license who may practice law, and the concern NPR reports bar officials raising is one of unauthorized practice, a status question about who may argue, independent of whether the AI's arguments would have been accurate. DoNotPay's own current homepage, as retrieved on 16 September 2026, describes the company as offering “AI-powered tools to take control, save time, and get what you deserve,” framing the product as consumer self-help rather than legal representation.

Where it fails

The documented failure is not technical — no transcript of a flawed AI argument exists, because the hearing never used the system. The open question the episode leaves is regulatory: courtroom recording rules, which NPR notes prohibit audio recording in federal court and often in state court, would have constrained the design regardless of the unauthorized-practice question, since generating a response required recording the proceeding.

• Does the jurisdiction permit recording the proceeding the tool would need to listen to?
• Is the activity considered practicing law under that state's definition, and does the tool operate under a licensed person's supervision?
• Was any independent testing of the tool's argument quality conducted in a lower-stakes setting first?

DoNotPay's own account frames the cancellation as a response to the risk of prosecution, not a concession that the technology was unready; bar officials frame their concern as one of licensure, not AI accuracy. Both framings are the parties' own, not a court's ruling.