If California sets AI-use rules, lawyers still face consequences for bad work

On October 1, 2026, California Governor Gavin Newsom signed into law first-of-its-kind state legislation setting rules for how lawyers are permitted to use generative AI in their work, according to Reuters Legal. Reuters Legal reported that Newsom signed the bill alongside a slate of other AI-related measures. Bloomberg Law, Business & Practice reported the same day that the measure covers lawyer and arbitrator AI use, while Legaltech News reported that SB 574 bars arbitrators from relinquishing their decision-making powers to generative artificial intelligence.

The significance for AI-native legal work is direct: the use of a generative system does not move responsibility out of the legal task. California’s legislation treats lawyer use as something subject to rules, not as an invisible production detail. Legaltech News describes stronger AI review responsibilities under the new California law, and Bloomberg Law says the law gives law firms the task of determining when their use of the technology requires disclosure. That places verification and accountability alongside the promised speed of AI-native delivery. A finished legal product remains exposed to consequences if the work is not adequately reviewed or if its production requires disclosure that is not made.

The event proves that California has enacted legislation regulating attorneys’ use of generative AI and that the legislation addresses both review responsibilities and disclosure decisions. It also establishes a separate limit for arbitrators: Legaltech News reports that SB 574 bars them from relinquishing decision-making powers to generative artificial intelligence. Those are concrete constraints on how legal work and adjudicative work may be performed in California. They do not prove that the law will catch every error, that every firm will disclose AI use in the same circumstances, or that review will reliably prevent defective work. They do not establish that AI-native firms are more accurate than conventional firms. The stronger reading would require observable application of the California law to actual work, including a disclosure decision, a finding that review was insufficient, or an order addressing an AI-related failure. None of those applications appears in the reported signing of the law.

The law therefore shifts the question from whether AI can produce legal work to whether the work can survive the obligations attached to its production. Bloomberg Law reports that the disclosure question is difficult for law firms, while Legaltech News reports the stronger review responsibilities. Difficulty cuts in both directions. It can make AI-native delivery more accountable by forcing firms to identify where human review remains necessary. It can also expose a weakness if a firm cannot explain what was reviewed, who made the legal judgment, or whether the final work required disclosure under California law. The legislation does not answer those operational questions. It makes them part of the conditions under which AI-assisted legal work is assessed.

The next confirming fact would be a California filing, order, or public law-firm disclosure applying the new requirements to a specific use of generative AI. A decision identifying inadequate review or imposing a consequence for an undisclosed use would strengthen the reading that AI does not eliminate the consequences of bad legal work. A clear disclosure and review practice without an enforcement dispute would show that the law is shaping production before a failure occurs, but it would not prove that the resulting legal work is accurate. We will measure the law by those observable applications, not by the existence of the technology or the signing ceremony.