Report from Bloomberg
In Brief – In response to concerns nationwide over AI-generated legal filings containing hallucinated or misinterpreted case citations, California has enacted first-in-the-nation legislation regulating how attorneys use generative artificial intelligence. SB 574 prohibits attorneys from delegating legal practice to AI, requires them to verify all citations before filing court briefs, including AI-generated citations, and requires at least one attorney to sign each filing in their individual name. Attorneys must also disclose when generative AI was used to create documents filed with courts and prohibits attorneys from entering confidential or personal information into open-access AI tools. The law applies similar restrictions to arbitrators, who cannot delegate decision-making to AI or rely on AI-generated information without disclosing it to parties. Courts may impose sanctions for violations.
Context – Chatbot “hallucinations” in legal briefs first made news in mid-2023, less than a year after ChatGPT’s release, with two lawyers being sanctioned in US District Court. Chief Justice John Roberts led off his 2023 Annual Report on the Federal Judiciary by discussing AI technology in the legal system, warning about hallucinations, and urged responsible use of the technology. Similar cases have continued to pop up, with lawyers facing various sanctions. The Judicial Office of the UK Ministry of Justice released AI guidance in late 2023 to judges and courts, warning about hallucinations and that US-centric AI tools might not UK legal tradition. At least one US judge who released an opinion riddled with likely hallucinations should have taken heed, with the fabricated citations finding their way into filings in at least one other case. This past May, the Florida Supreme Court released an update to the state’s general practice rules that were focused on AI, requiring the signer of any document filed with Florida courts to ensure that “the legal authorities identified exist and are accurately cited.” Of course, everyone should know by now that they need to check all AI work. This is not complicated, just hard to do.