Joe Stephens
Director of Legal Solutions
Steno
Steno focuses on providing attorneys with innovative tools and options that overcome the technological and financial hurdles that arise when proving a case. As a partner in the pursuit of justice, Steno understands the unique challenges faced by legal professionals and is committed to delivering practical, impactful solutions.
I built a rural public defender office from scratch. Today, I run the only public defender office in the country housed inside a law school. When generative artificial intelligence first emerged, it seemed like a revolutionary answer for offices like mine: a way to relieve a permanent, systemic crisis in criminal indigent defense. Public defenders routinely carry caseloads that no honest reading of the Sixth Amendment can justify, forced to triage justice because there are simply never enough hours in a day.
The initial theory of AI in the access to justice space was attractive: it would absorb the routine administrative drag. It could draft boilerplate motions, organize chaotic medical and police records, surface hidden constitutional issues, and free attorneys, like my team, to spend our limited time where it actually matters--with our clients, in jail and in the courtroom. We were starting to act on that promise, using these systems to close our representation gap. Then came the regulators.
Regulatory panic
In March 2026, the California State Bar's Standing Committee on Professional Responsibility and Conduct (COPRAC) put forward a package of proposed amendments designed to turn AI guidance into enforceable disciplinary rules. Because California licenses more attorneys than any other state, its regulatory choices rarely stay local. If other state bars pursue this exact rulemaking route, the result will not just be a baseline standard for technological competence; it will be the practical elimination of AI as a tool for expanding access to justice.
The friction centers on COPRAC's proposed comment to Rule 1.1, governing competence. The proposed draft states that when using technology, a lawyer "must independently review, verify, and exercise professional judgment regarding any output generated by the technology that is used in connection with representing a client." And because the judgment the rule demands is the lawyer's own, it cannot be handed to a paralegal or other nonlawyer. The proposed update to Rule 5.3 reinforces the point, treating supervision of nonlawyer AI use as part of the lawyer's nondelegable duty.
Consider what this does to the operational math of an overburdened legal aid or public defender office. If an attorney carrying 400 active files uses an advanced AI tool to draft a standard motion to suppress, the rule requires that the attorney personally audit every line or fact, before filing.
That means the operational bottleneck does not disappear; it moves. Instead of spending an hour drafting from a template, the lawyer now spends that hour line editing unfamiliar text, and for many filings the careful review of someone else's draft demands as much attention as writing it from scratch. The gains do not vanish everywhere. They vanish in the one place an overloaded office needs them most, the high-volume routine work, because the rule refuses to let scrutiny scale to the stakes. The same exacting audit is owed to a form answer and to a dispositive brief. The efficiency that was supposed to let us represent the unrepresented gets spent instead on a standard that treats every document as equally dangerous.
For pro se or self-represented litigants, the impact of these proposed changes is even more severe. The grandest public interest vision for legal technology has always been direct-to-consumer AI tools that help low-income individuals navigate complex court systems without the expense of hiring an attorney. Under the proposed California framework, if an unrepresented person uses a standalone AI tool to generate an answer to an eviction notice, no supervising lawyer is present to fulfill the independent verification requirement.
State bars pursuing this path leave only two realistic options for public-facing legal tools. Either these platforms must be restricted entirely to lawyer-mediated use, which defeats the purpose of expanding access, or they will be forced to operate entirely outside the regulated profession. This pushes vital resources back into the legal gray zone of unmonitored form-filling services, leaving vulnerable litigants with fewer protections, not more.
Who does this benefit?
This dynamic forces a difficult question that the legal profession routinely avoids: Are these rules designed to protect clients from technological mistakes, or to protect the economic monopoly of the bar from technological competition?
The answer is likely a mix of both. The risks of AI hallucinations and inaccurate citations are well-documented and real. But by failing to distinguish between types of legal work, the rule applies a blunt disciplinary standard that falls heaviest on the populations who already cannot afford counsel. A corporate firm billing $1,000 an hour can pass the cost of manual, multi-layered associate review onto an institutional client. A rural public defender or a legal aid lawyer protecting a tenant from immediate homelessness cannot.
An alternative path
A more rational regulatory framework would not treat every output as identical in terms of disciplinary risk. First, rulemakers should set tiered verification standards calibrated to the stakes and the use. A chronology pulled from a client's medical records and checked against the source does not carry the risk of a brief filed under a lawyer's signature, and the rule should not pretend it does.
Second, rules must clearly distinguish between enterprise tools used by lawyers and consumer tools provided directly to the public, relying on clear disclosures rather than outright prohibition.
Third, regulations should recognize that automated, system-level quality assurance can reliably substitute for manual human review in defined, low-risk contexts.
Finally, state bars must build explicit carve-outs for legal aid and indigent defense offices that weigh the marginal risk of a technological error against the absolute certainty of zero representation. These changes would let AI support access to justice without sacrificing accountability.
We can look at how other fields with real stakes manage this. Medicine did not meet clinical decision support by ordering physicians to recompute every value a diagnostic system produces. It built oversight that scales to risk. A flagged anomaly on a scan draws a different level of human scrutiny than a routine measurement, and the heaviest controls, including FDA device review, attach to the highest risk uses rather than to all uses at once. That is the distinction the proposed rule refuses to draw.
The legal profession is moving toward a framework that treats a routine administrative draft and a capital murder brief as requiring the same level of personal attorney review. No other advanced profession handles technological transition this way. If we continue down this path, we will preserve the traditional mechanics of legal practice at the direct expense of the people the justice system is meant to serve. Rulemakers should change course now and adopt standards that protect clients without shutting down access to justice.