Uncontrolled Use of AI – court guidelines 

Uncontrolled Use of AI – court guidelines
Ronit Warshai, Adv. 
Levitan, Sharon & Co.

The use of AI programs such as ChatGPT and similar tools has rapidly become a popular research resource for legal matters. The legal system consists of vast amounts of written material, including statutes, regulations, judicial decisions, and academic articles, all of which AI systems can readily access and summarize. However, this benefit also has a darker side.

One of the well-known characteristics of generative AI systems is their tendency to provide answers that satisfy the user, even when those answers are not based on reliable information. In some cases, AI programs invent statutes, regulations, case law, or quotations. This phenomenon is commonly known as “hallucination.” Courts have already encountered situations in which parties cited non-existent laws or precedents allegedly supporting their arguments.

The following case illustrates the consequences of relying on unverified AI-generated content, the negligence involved in failing to verify such information, and the resulting question of liability.

C.A. 63194-08-25 – Nevo Ben Cohen v. Ramat Gan Municipality & Others

In March 2026, the Israeli Supreme Court delivered its judgment in a case that reads almost like a practical guide to what not to do when using generative AI in legal and administrative work.

The dispute began as a routine administrative matter. Ramat Gan Municipality denied a request for special transportation services from the residence of a divorced father, relying on what it claimed was a Ministry of Education Director-General Circular and on “recent case law” allegedly interpreting that circular. The problem was straightforward: the circular did not exist, and several of the cited judicial decisions and quotations were entirely fabricated.

The father contacted the Ministry of Education and requested a copy of the alleged circular. The Ministry responded that it was unaware of any such circular. The father then approached the Municipality with the Ministry’s response but received no reply.

The father subsequently filed an administrative petition with the District Court. Even in its response to the court, the Municipality and/or its counsel continued to rely on the fictitious circular and the non-existent precedents.

On the substantive issue of transportation services, the parties eventually reached a settlement. Nevertheless, the father argued that, in light of the Municipality’s bad-faith conduct in repeatedly relying on AI-generated content without verification, costs should be awarded against it.

The District Court denied the father’s request for costs, leading to an appeal before the Supreme Court.

The Supreme Court held that the Municipality’s unrestrained reliance on artificial intelligence without reviewing or verifying the information generated amounted to recklessness, constituted serious misconduct, and was wholly unacceptable, particularly given the heightened duty of fairness owed by an administrative authority.

Accordingly, the Court ruled that costs should be imposed on the Municipality at the higher end of the scale.

This judgment extends far beyond the field of municipal law. It represents a clear judicial statement regarding accountability in the age of artificial intelligence and has implications for all users of AI-generated content.

The Court’s Core Message: AI Is Not the Problem – Unverified AI Output Is

The Supreme Court drew an important distinction. The use of AI tools is not inherently improper. What is improper is relying on AI-generated content without adequate oversight, verification, and human judgment when that content forms the basis of decisions, legal submissions, or communications with the public.

In this case, the citizen independently examined the alleged circular and the cited case law and discovered that neither existed. Despite being informed of these inaccuracies, the Municipality failed to correct its position until formal legal proceedings were initiated.

The Court regarded the submission of pleadings containing unverified AI-generated content as a serious breach of professional obligations and a severe misrepresentation to both the court and the opposing party. Importantly, the Municipality’s conduct was not an isolated mistake; the inaccuracies were repeated at several stages of the proceedings, including after they had been explicitly brought to their attention.

This is not the first occasion on which the Supreme Court has addressed the improper use of AI-generated material in legal proceedings.

What makes this case unique is the Municipality’s persistent reliance on fabricated information in its dealings with a private citizen.

The Court emphasized that individual citizens often lack the expertise, resources, or confidence necessary to challenge official communication that appears well-reasoned and legally supported. A decision that appears lawful on its face may rest on entirely fictitious foundations and may never be subjected to meaningful review. The Court therefore concluded that the Municipality’s conduct amounted to serious negligence.

The Court also rejected the Municipality’s attempt to shift responsibility to its attorney by arguing that counsel had failed to correct the submissions after the substantive dispute had been resolved.

Important lessons can be drawn from this and other cases involving AI-generated content. Such material should always be reviewed and verified with the same degree of care that would be applied to work prepared by a junior associate or trainee. AI can be a valuable tool, but it cannot replace professional judgment, independent verification, and accountability.