Israeli Bar Association Bars AI Agents from Making Legal Decisions
The Israel Bar Association's National Ethics Committee has issued a strict directive prohibiting lawyers from allowing autonomous systems or AI agents to make substantive legal decisions, provide independent legal advice, draft legal documents, or advance legal proceedings without close human supervision and approval. The updated guidance, signed by Committee Chairman Adv. Menachem Moshkovitz, clarifies that while AI can serve as a controlled technical aid, the ultimate legal and professional responsibility rests solely with the human lawyer, emphasizing that human judgment is irreplaceable.
AI systems, including increasingly autonomous 'AI Agents' capable of planning work, selecting tools, and interacting with external systems without human intervention, pose growing ethical, professional, and procedural risks as their autonomy increases. The directive states that AI systems must not be used to make decisions on behalf of a lawyer regarding representation, provide independent legal advice, perform representation actions towards clients, courts, or third parties, file legal documents, serve binding documents, or advance legal proceedings without sufficient human oversight and approval.
However, the committee permits the use of AI for specific, documented, and reversible technical tasks that are based on prior human decisions and are subject to clear parameters, appropriate controls, and human supervision. The presence of an autonomous component or AI agent does not inherently ban its use, but requires lawyers to carefully assess its suitability, the nature of its actions, and the level of human oversight.
Lawyers remain personally and directly responsible for all professional actions, decisions, advice, arguments, or outputs generated using AI. They must protect client information, including confidential and personal data, from unauthorized disclosure when using AI systems, avoiding any use that could compromise privacy or breach confidentiality obligations. The directive specifically warns against inputting sensitive or non-public information into unsecured, open AI systems, even after attempts to anonymize data, as cross-referencing could still reveal identities.
Transparency with clients is generally not required for routine, technical AI use that doesn't involve open platforms or limited information. However, disclosure is necessary when AI use significantly impacts representation goals, involves inputting limited information into open AI platforms, substantially affects professional decision-making, is requested by the client, conflicts with client instructions, or affects fees or expenses. Obtaining explicit, informed consent is required before inputting limited client information into open AI platforms, especially if anonymization is insufficient to prevent identification.
The Bar Association also warns against AI 'hallucinations,' where systems fabricate plausible but incorrect legal citations or rulings. Lawyers relying blindly on AI outputs that contain such fabrications will face sanctions, reputational damage, and disciplinary action. While AI can offer insights, drafting assistance, and initial work, it cannot replace human professional judgment, factual investigation, and decision-making. The committee stressed that failure to comply with the guidance could be considered evidence of ethical breaches, while adherence would indicate ethical and reasonable conduct but not absolve lawyers of their duty to act according to the law and exercise individual judgment.
