Bad legal work still has consequences if California’s AI law requires review

On October 1, California Governor Gavin Newsom signed into law legislation setting rules for how lawyers may use generative artificial intelligence in their work, according to Reuters Legal. Bloomberg Law, Business & Practice reports that the measure is the first of its kind. The law also reaches arbitration: Legaltech News (Law.com) reports that SB 574 bars arbitrators from relinquishing their decision-making powers to generative artificial intelligence.

That matters because the law treats AI use as a problem of professional responsibility, not as a transfer of responsibility to a tool. Legaltech News (Law.com) reports that the measure gives lawyers stronger AI review responsibilities. Bloomberg Law says California’s law gives firms the task of deciding when their use of the technology requires disclosure. Those duties do not establish that every use produces bad legal work, but they do establish that automation does not by itself settle who remains answerable for the work delivered to a client or presented in a proceeding. For an AI-native firm, the relevant promise is therefore not merely that generative artificial intelligence can produce legal work. The promise has to survive review, disclosure decisions and retained human decision-making under California law.

The law proves that California has imposed rules around lawyers’ use of generative artificial intelligence and that California has kept review and decision-making responsibilities in view. It does not prove that the rules will prevent inaccurate filings, defective advice or other bad legal work. It also does not prove that a firm’s internal review is effective, that disclosure will reveal every material use, or that a client will receive work at lower cost or higher quality. Bloomberg Law’s description of tough disclosure decisions points to a compliance question, not an outcome measure. The stronger reading would require a concrete matter showing how a lawyer or firm used the technology, what review occurred, what work reached a client or tribunal, and what consequence followed. Nothing in the reported signing establishes those facts for any particular matter. The order of accountability remains intact; the performance of any particular AI-enabled legal provider remains untested here.

The next confirming fact would be a California proceeding applying the new rules to a specific lawyer or firm: an order identifying the AI use, the review performed, any required disclosure and the consequence for the resulting legal work. A filing or order that links deficient work to inadequate review would show the statute operating as an accountability rule rather than a disclosure exercise. A matter in which disclosure occurs but bad work produces no identified consequence would cut against the stronger practical reading. Until such a proceeding appears, the statute supplies a governing obligation in California, not proof that any AI-native delivery model has made bad legal work consequence-free. We will score the model on the completed matter, not the existence of the rule.