A professional body fills the space the state left open
In late September 2026, the National Ethics Committee of the Israel Bar Association issued a second, expanded opinion on how lawyers may use AI. Pearl Cohen's write-up is dated 29 September 2026. It updates the Committee's preliminary guidance of May 2024. We could not confirm the exact date the Committee issued it. Pearl Cohen says the Committee acted because AI is developing fast, members keep asking questions, and Israeli courts have sanctioned lawyers for improper AI use.
The opinion's core is simple. AI is, in Pearl Cohen's words, an auxiliary tool and not a legal entity, and a lawyer "may not hide behind AI outputs or rely on them blindly." The Hebrew text, hosted by the Bar, says the lawyer bears personal and direct responsibility for every act, advice, argument or work product produced with AI. It also says AI does not replace the lawyer's duty of independent professional judgment.
What the opinion actually requires
Three obligations stand out, based on Pearl Cohen's summary and the Hebrew text.
- Verification. Every AI output must be verified before the lawyer relies on it, presents it, or sends it to a client or court. Pearl Cohen reports that the Committee cites a Supreme Court ruling that placed the fault in uncontrolled use without proper verification and review.
- Agents. For agentic AI, every action of legal, procedural or professional significance must be reviewed and approved by a human. Autonomous agents do not make substantive legal decisions.
- Confidentiality. Sharing restricted client information with an external AI provider may be an impermissible waiver of confidentiality or privilege. Lawyers must check the provider's terms, storage and deletion practices. Where risk remains, they must inform the client and obtain informed consent.
This is a tighter version of the May 2024 guidance. Pearl Cohen's 2024 summary shows that the earlier opinion already urged verification, caution with identifying client details, and written consent for open systems.
The strongest case for more binding rules
The best argument for statute over professional guidance is that hallucinated citations and leaked client data are not only a lawyer's problem. Courts, opposing parties and clients bear the cost, and an ethics opinion binds only Bar members. Litigants who represent themselves, or AI products sold directly to the public, sit outside the Bar's reach. A single binding rulebook would also give regulated industries certainty.
That case is serious, but the opinion shows why a narrower tool works better here. The duties it creates fall on the party best placed to carry them: the licensed professional who signs the filing. They are technology-neutral. "Verify before you rely" applies to today's chatbots and to whatever replaces them. It requires no new regulator, no licensing of models and no compliance regime that a small Tel Aviv legal-tech startup would struggle to afford. It also imposes accountability where harm actually occurs, in the use of outputs, rather than on research and development.
The state's own AI framework is the weaker link
Israel's national posture is soft law. The Legal 500 Israel AI guide states that "no binding AI regulations have been formally enacted in Israel" and that there is no dedicated, AI-specific statute. It describes the December 2023 policy from the Ministry of Innovation, Science and Technology and the Ministry of Justice as a sector-specific approach.
We think that approach is right in principle. Sector regulators and professional bodies know their own risks better than a horizontal AI act would. The weakness is execution. The State Comptroller's November 2024 special report on national preparedness for AI found that, six years after the Prime Minister's decision, a comprehensive national plan had not been approved by the government. It also found no coordinating government entity with overall responsibility. It noted that the principles in the national AI policy document had yet to receive government approval. It added that only about 40% of the Phase I budget had been used: NIS 220 million of NIS 550 million.
The contrast is instructive. The Bar, a body with a clear mandate and a clear subject, produced a usable, updated rule in about two years. The state's strategy, as of the Comptroller's audit, was still waiting for approval. Soft law only works if someone owns it.
What follows from this
Light-touch regulation is a choice that has to be run well. Three things would make Israel's approach credible without importing a heavy EU-style regime.
- Approve and own the policy. The government should formally adopt the principles and name a coordinating body, as the Comptroller recommended.
- Let professions lead. Other regulated professions, such as medicine, accounting and finance, can adapt the Bar's verification-plus-consent model to their own settings.
- Keep duties on use, not on models. Rules that attach to a professional's reliance on outputs stay useful as the technology changes, and they do not push AI builders out of the country.
The Bar's opinion is not a model for the whole economy, but it shows what proportionate rules look like: clear responsibility, a verification duty, and informed consent where confidential data leaves the building. The open question is whether the government will match that discipline.