ABA 关于生成式人工智能的正式意见 512

Opinion 512 does not create new rules. It explains how the ABA Model Rules of Professional Conduct already apply when lawyers use generative AI. It is advisory: each state adopts and interprets its own rules, and some, including California and Florida, issued guidance before the ABA did. Competence (Rule 1.1). Lawyers need a reasonable understanding of the capabilities and limitations of the specific tools they use, not expertise in machine learning. Because output can be inaccurate, the right amount of independent verification depends on the task and the tool, and uncritical reliance is not acceptable. Confidentiality (Rule 1.6). Before inputting information relating to a representation, lawyers must assess the risk of disclosure or access, including by others inside the firm who use the same tool. For self-learning tools that may use inputs to shape later outputs, the opinion says the client's informed consent is required, and boilerplate consent in an engagement letter is not enough. Communication (Rule 1.4). Not every use must be disclosed, but lawyers must inform clients when asked, when client information will go into such a tool, or when AI use bears on decisions the client makes, such as the basis of the fee. Candor and meritorious claims (Rules 3.1, 3.3 and 8.4(c)). Lawyers must check AI-generated citations and analysis before submitting them to a tribunal. Supervision (Rules 5.1 and 5.3). Managerial lawyers should set clear policies, train lawyers and staff, and vet vendors. Fees (Rule 1.5). Hourly bills must reflect time actually spent, including review. Lawyers generally may not bill for learning a tool they will use broadly. General-purpose tools are usually overhead, while direct costs of a tool used for a specific matter may be charged if reasonable and disclosed. A frequent misconception is that the opinion bans consumer chatbots. It does not; it requires informed judgment about each use.