Court ruling under scrutiny after AI-linked legal citations found to be inaccurate
Distorted quotations from Israeli law raise questions about the role of AI in judicial drafting.
A ruling in a small claims court case contains fabricated or distorted quotations from Israeli legislation, possibly as a result of the senior registrar’s use of artificial intelligence (AI) in the drafting process.
In recent months, Israeli courts have repeatedly been forced to address the use of AI by lawyers in drafting pleadings that included fabricated arguments and references. In February, in a High Court ruling, Justice Gila Canfy-Steinitz addressed the issue after it emerged that a petition had been based on fabricated judgments or citations generated with the help of AI. “The use of artificial intelligence tools by lawyers does not absolve them of their professional responsibility and judgment,” Canfט-Steinitz wrote in her decision rejecting the petition. A few days later, Supreme Court Justice Noam Sohlberg dismissed a similar petition outright. “There is concern that we are dealing with a phenomenon, or a troubling wave,” he wrote. “It must be made clear to all concerned what is expected of them when they use artificial intelligence. There are certainly benefits to this tool, but one must beware of its pitfalls.” In March, Ramat Gan Municipality was ordered to pay NIS 30,000 after the Supreme Court found that it had relied on fabricated AI-generated quotes in its pleadings.
Now, it appears that the court itself may also have made improper use of AI. In January, Senior Registrar Michael Shempel issued a ruling in a lawsuit under the Spam Law filed with the Tel Aviv-Yafo Small Claims Court. The ruling contains several citations of legislation that are distorted or do not exist, with indications suggesting careless or incorrect use of AI in the drafting process.
For example, the ruling states that Section 30A(e) of the Communications Law provides that “a recipient may at any time notify the advertiser that they refuse to receive advertising material.” However, the actual wording of the law states that “an advertiser who sends advertising material under this section must include certain details in a clear and prominent manner so as not to mislead,” followed by additional required specifications. These elements are omitted or inaccurately reproduced in the ruling, and are used by the registrar to conclude that the defendant violated the law.
In another instance, the ruling defines advertising as “a message distributed for commercial purposes, intended, directly or indirectly, to encourage the purchase of a product or service or expenditure of funds in another manner,” although the statutory definition does not include the phrase “directly or indirectly.”
The ruling also cites Section 30A(b) as stating that “an advertiser shall not send advertising by fax, automatic dialing system, electronic message, or SMS without the recipient’s prior express consent, including by electronic message or recorded conversation; however, the advertiser may contact the recipient, as determined by the Minister, to obtain such consent.” In fact, the law includes detailed provisions regarding consent and exemptions but does not contain a clause granting discretionary authority to the Minister in the manner described in the ruling.
In March, the Courts Administration published a code of ethics for the use of AI by judges, registrars, and court staff. The code, issued after the ruling in question, states that judicial officers must exercise independent judgment when using AI, that its use should be appropriately disclosed, and that users must “verify every factual determination, quotation, reference to legislation and case law, legal summary, and any other information against authoritative sources.”
The Courts spokesperson responded: “There were indeed inaccuracies in the citation of statutory provisions, and this is regrettable. However, these inaccuracies do not affect the outcome of the ruling. The decision may be appealed in accordance with the law.”
