AI for lawyers in Israel: what the Bar Association allows, where it saves time and where it is risky

AI for lawyers is no longer a question of "whether", but of "how, and within what limits". On 17 September 2026 an updated ethics opinion was published by the National Ethics Committee of the Israel Bar Association, tightening the rules for the use of artificial intelligence, and of AI agents in particular. In this article we explain what it says, what must never be entered into an AI tool, when client consent is required, and why every citation of case law has to be checked. We finish with where AI genuinely saves time in a law office. One thing to say up front: we are a marketing agency, not lawyers, and this is not legal advice.

What the Bar Association's new ethics opinion says

According to the report in ynet of 17 September 2026, the opinion is signed by the chair of the National Ethics Committee, Adv. Menachem Moshkovitz. The central rule, as quoted there: "No use shall be made of autonomous systems or AI agents in a manner that allows them to take, in the lawyer's place, substantive decisions concerning the representation, to give independent legal advice, to perform acts of representation towards a client, a court or a third party, to file pleadings, to effect service of binding documents, or to advance a legal step without sufficient human supervision and approval."

And what is allowed? According to the same report, there is nothing to prevent the use of a system for actions that are "technical, bounded, documented and reversible", carried out on the basis of a prior human decision, within clear parameters and under human supervision.

Three further points from the document, as published:

• Personal responsibility. A lawyer "may not hide behind the outputs of the artificial intelligence".
• A prima facie presumption. Failure to comply with the opinion "may constitute a prima facie presumption, albeit a rebuttable one" that the lawyer has not met their ethical duties.
• A continuation of the earlier opinion. This is the second opinion on the subject. The first, a preliminary ethics opinion (decision את/60/24) of 7 May 2024, already held that the client's consent must be obtained in advance for the use of "open" AT software, and noted that the concerns exist in closed in-house systems too.

Note: we read the wording of the updated opinion as quoted in ynet. Before you change procedures in the office, read the Bar Association's full document.

Confidentiality: what must never be entered into an AI tool

The opinion defines "restricted information": any information that is not public, that was given to the lawyer and that is subject to a duty to limit its disclosure. According to the document, such information must not be entered into "open AI platforms, or any other environment about which there is insufficient certainty that the information will remain protected at a level matching the lawyer's professional duties".

Three things that are easy to miss:

1. It is not only about documents. The rule also applies to "partial items of information, case descriptions, email drafts, commercial data, medical information, distinctive facts". In other words, "draft me an email to the other side's lawyer" with the details of the file counts too.
2. Simple de-identification is not enough. According to the Bar Association, removing the client's name or an identifying detail is not sufficient protection, because cross-referencing data may expose the identity.
3. A closed system is not an exemption either. You have to examine "how the data is stored, the permissions, the logs, the encryption", and make sure that anyone not handling the file cannot see the information — inside the office as well.

What to check with an AI tool supplier, according to the requirements set out in the opinion:

• That the supplier undertakes not to share the information with third parties.
• That the information is not used to train or improve the model without appropriate consent.
• That the information and the processing outputs are not included in external databases outside the firm's control.
• That there is information security, access control, deletion and data retention at a reasonable standard.

The privacy-law side of this is covered in our guide to Amendment 13 for businesses.

When you must tell the client, and when you need consent

According to the opinion, the mere use of an AI tool does not require disclosure in every case, particularly in routine or technical use. Transparency towards the client is required as a rule, however, when:

• the use relates substantively to the way in which the goals of the representation are achieved;
• you wish to enter restricted information into an open platform;
• the use is likely to have a substantive effect on professional decision-making;
• the client has asked, or has given instructions of their own about AI;
• the use is relevant to the fee or to expenses.

And what about consent? When a client's restricted information is used on an open platform, even after de-identification, the document requires "the client's informed and explicit consent in advance", if the de-identification does not sufficiently rule out the possibility of identifying them. In addition, a general clause in the fee agreement or the power of attorney stating that the lawyer "may use artificial intelligence" is not, as a rule, sufficient as informed consent.

The practical conclusion: a general sentence about AI in the engagement agreement does not settle the matter.

Invented case law: why every citation has to be checked

This risk has already reached the Supreme Court. According to a report on ICE from February 2025, a petition filed with the Supreme Court relied on 36 judgments. On examination, five of them did not exist at all; on top of that, 14 of the references contained basic errors, and 24 of the quotations taken from them were not found in the source. Justice Gila Kanfi-Steinitz imposed no sanction, because it was the first case of its kind, but warned that in future cases the courts may impose sanctions, including striking out the proceeding in limine and personal costs against the lawyers.

The new opinion explains why this happens: the systems "are not necessarily designed to provide information that is factually correct or legally accurate", and often produce an answer that is "statistically correct" and looks convincing.

A simple checking routine for every draft:

1. Every judgment the tool mentioned: open it in a case-law database, not through the link the tool gave you.
2. Every quotation: search for the exact sentence in the original text.
3. Check that the ruling is still good law and has not been overturned on appeal.
4. Mark on the draft what was checked and who checked it, so that the check itself is documented.

Where AI for lawyers really saves time, and the risk in each use

The table below is our reading of the opinion as published, as a starting point for a conversation in the office. It is not legal advice.

In our estimate, the biggest time saving from AI for lawyers in a small office is not in drafting pleadings. It is around the legal work: first responses to enquiries, scheduling meetings, reminders, summarising public material and content for the website. Those are precisely the areas that do not touch restricted information. For everyday uses of tools like ChatGPT, see our guide to ChatGPT for business.

An AI agent in the office? Before you connect one, answer four questions in writing: which systems it has access to, which actions it performs on its own, what always waits for a lawyer's approval, and where the action log is kept. What an AI agent is and how it works is explained in our guide to AI agents for business.

What Simple Web does about it

Simple Web is an AI-first marketing agency from Bnei Brak, certified Meta partners and Google advertising experts, working with 200+ businesses and holding 28 five-star Google reviews. We are not lawyers, and the decision about what is permitted in the office belongs to the lawyer.

We build AI agents for business that work inside the business: customer service, calendar and email, and reporting on WhatsApp, Telegram or email. Every agent has defined permissions and a full action log, and sensitive actions require human approval. The information stays in your systems, and the work is done in line with Israeli privacy law, including Amendment 13. From the spec meeting to a first agent live takes 2–4 weeks, and the price is set in a quote based on scope.

For a lawyer who wants personal help, there is a personal AI assistant that connects to the calendar, email and tasks. Sending an email, booking a meeting with other people or making a payment all wait for your approval, and any account can be disconnected at any moment. It is ready within two weeks of the introductory call, with a quote based on scope. We have also built a brand website for Or Attel, Advocates and Mediation. What a law firm's website needs to include is set out in our guide to a website for a lawyer.

Summary

The September 2026 ethics opinion does not ban AI for lawyers. It establishes that the tool remains a tool: the lawyer decides, checks and bears responsibility. Restricted information does not go into open platforms without informed consent, and a general sentence in an agreement is no substitute for it. Every judgment an AI tool cites is checked at source. Want to see where AI could save time in your office without touching client files? Book a free diagnostic call.

Sources

• ynet: the updated ethics opinion of the National Ethics Committee: the quotations from the opinion of 17 September 2026: AI agents, restricted information, informed consent and transparency
• Israel Bar Association: preliminary ethics opinion, decision את/60/24: the first opinion, of 7 May 2024, and the requirement of client consent for "open" software
• ICE: the Supreme Court ruling on invented case law: the February 2025 case and the warning about sanctions