Written by: Haim Ravia, Dotan Hammer
The National Ethics Committee of the Israel Bar Association issued a new ethics opinion on the use of artificial intelligence tools by lawyers. It is the second opinion on the subject, updating and expanding the preliminary opinion from May 2024. The Committee explains that it revisited the subject roughly two years on because of the rapid and extensive development of AI technologies, including dedicated tools for lawyers, and because of the volume of enquiries from members — noting that Israeli courts have already imposed sanctions on lawyers who made improper use of AI platforms in the course of legal representation.
The opinion is based on a set of defined terms. An “AI output” is any output obtained from an AI platform as a result of a query, instruction, prompt, document upload or other input, and may be a legal document, a draft agreement, a letter, a summary, a legal review, a case-law survey, an analysis, an image, an audio or video clip, or code. “Restricted information” is any non-public information which, by virtue of its nature, source or context, is subject to a legal, professional or ethical duty to protect it from disclosure to third parties; comprising personal information of a client or information permitting the client’s direct or indirect identification (as defined in section 3 of the Protection of Privacy Law as amended by Amendment 13); information protected by the duty of confidentiality or by attorney-client privilege; and non-public information relating to third parties whose disclosure could infringe their privacy or breach a duty owed to them.
The opinion also defines “anonymization,” with the express caveat that anonymization is neither absolute nor dichotomous: even after removal of a name, identity number or other identifier, re-identification may remain possible depending on the type of information, the combination of data points, information available from other sources and the reasonable means available to the user or to third parties.
On professional competence, the Committee states the governing principle: an AI system is an auxiliary tool and not a legal entity, and although it can produce output that appears to be a legal work product, that capability cannot replace — even in part — the lawyer’s independent, current professional judgment. The lawyer’s personal responsibility for actions, advice, legal documents, opinions and pleadings remains complete, and the lawyer may not hide behind AI outputs or rely on them blindly.
Competence has three components: familiarity with the opinion and with the guiding case law on the use of AI in legal practice; familiarity with the general characteristics of AI systems, the risks they create and accepted means of addressing them; and familiarity with the specific system in use — how it operates, the tasks for which it is suitable, its limitations, its terms of use, and the legal and professional risks arising from it, including bias, hallucinations, factual and legal errors, breaches of confidentiality and inaccuracies presented as reliable content. This competence is not a one-time obligation, and lawyers must develop a basic AI literacy sufficient to distinguish appropriate from dangerous or improper uses.
The obligation is amplified for systems with greater autonomy, including AI agents, which may plan workflows, select tools, propose or perform actions and advance tasks without immediate human intervention. The Committee requires that any use of an AI system, and in particular a system with autonomous characteristics, be conducted so that professional control and substantive legal decisions remain with the lawyer, and that every action of legal, procedural or professional significance be subject to review and approval by appropriate human actors.
On verification, the Committee is categorical: a lawyer is under a professional and ethical duty to verify every AI output before relying on it, incorporating it into professional work, transmitting it to a client, presenting it to a court or using it to any third party. That duty flows from the lawyer’s own responsibility for professionalism, accuracy, reliability and integrity, and is not diminished because the content was generated or drafted with technological assistance. Retrieval-based systems that draw on defined databases and cite specific sources may reduce the risk of error, but do not displace the independent duty of verification. The Committee cites the Supreme Court’s ruling in the Ramat Gan case, where the Court held that the fault lies not in the use of AI as such but in its uncontrolled use without appropriate verification and review. Responsibility for the quality of the work product remains with the professional, who does not discharge that duty by relying on another — whether a human being or a machine.
On confidentiality, privilege, privacy and information security, the Committee warns that sharing restricted information with an external AI provider may, depending on the circumstances, be construed as an impermissible waiver of the client’s confidentiality or privilege, or at least create a real risk of such a finding. Entering legal, factual or strategic information into an open AI system – in particular a large language model operated by an external provider – may negate the confidential character of the information or undermine a claim to attorney-client privilege, because the information has been disclosed to a third party that is not a lawyer and is not subject to an equivalent professional duty.
Before using such a system, the lawyer must examine with care the provider’s status, the terms of use, the scope of access to the information, how it is stored and deleted, and whether adequate contractual and technical protections exist; and where a risk exists, must inform the client and obtain informed consent.
Click here to read the National Ethics Committee’s opinion (in Hebrew).
