California's bar regulators have spent the past year building an ethical framework for AI in legal practice. The framework has a problem: it doesn't account for how the most consequential AI tools actually work.
Writing for the Daily Journal, Joe Stephens, JD, Steno's Director of Legal Solutions, examines the draft rules proposed by California's Standing Committee on Professional Responsibility and Conduct (COPRAC) and the gap at their center. The California Supreme Court specifically asked COPRAC to address agentic AI—systems that can plan, reason, and carry out complex multi-step tasks with little human involvement along the way. The word "agentic" doesn't appear in the draft.
That omission was deliberate. COPRAC chose to fold AI into existing professional conduct rules rather than create a standalone framework, and that choice creates a structural conflict. The existing rules were written for tools that respond to a single prompt and produce a single output under direct supervision. They don't translate cleanly to systems that operate autonomously across many steps.
The verification and supervision requirements in the draft either become impossible to satisfy in practice, or they strip out the efficiency that makes agentic tools worth deploying.
The stakes extend beyond California. With more licensed attorneys than any other state, California's ethics rules carry gravitational pull across the profession. If other bars adopt the same approach, the result may not be an explicit ban on agentic AI but could create an effective prohibition through the economics of compliance.
Joe maps three likely outcomes if the current draft moves forward as written, and makes the case for a more precise policy response before the window closes.
Read the full piece in the Daily Journal.