Using AI Ethically in Law: Duties and Safeguards

For most professions, using AI well is a matter of productivity and quality. For lawyers, it is also a matter of professional ethics, because a lawyer's conduct is governed by binding rules of professional responsibility, and AI use touches several of them directly.

This is why the previous lessons kept returning to verification and confidentiality: they are not just good habits but expressions of ethical duties that carry real consequences, discipline, sanctions, malpractice liability, if breached. Understanding AI as an ethics issue, not merely a tech issue, is what separates safe adoption from dangerous adoption.

Bar authorities have recognized this. The American Bar Association issued Formal Opinion 512 in 2024, addressing how lawyers' ethical duties apply to generative AI, and many state bars and courts have issued their own guidance and rules. The clear message across them is that existing professional duties fully apply to AI use, no new exception, and lawyers must understand the technology well enough to use it consistently with those duties.

This lesson maps AI use onto the core duties: competence (understanding the tools), confidentiality (protecting client information), candor and verification (not relying on or filing false output), and reasonable fees (billing honestly for AI-assisted work). Then it turns those duties into a practical responsible-use policy. The goal is to let you capture AI's benefits while staying firmly within your professional obligations, because in law, how you use a tool is itself subject to the rules.

The foundational duty is competence: a lawyer must provide competent representation, and in the modern understanding that includes technological competence, keeping reasonably abreast of the benefits and risks of relevant technology.

Applied to AI, competence has two sides. First, a lawyer should understand AI tools well enough to use them effectively where they genuinely help, since refusing all useful technology can itself be a disservice to clients who deserve efficient, affordable representation. Second, and more pressingly, a lawyer must understand AI's limitations well enough to use it safely, knowing that it can hallucinate, that it can breach confidentiality, and that its output requires verification.

The competence duty is precisely why a lawyer cannot treat AI as a magic black box. You do not need to be an engineer, but you must understand, at a working level:

• what the tool can and cannot reliably do,
• how it can fail (hallucination, bias, outdated information),
• what happens to data you put into it,
• and therefore what checks its output requires.

This reframes the entire cursus. The verification and confidentiality practices from the earlier lessons are not optional best practices; they flow from the ethical duty of competence. A lawyer who files AI-hallucinated citations has not just made a mistake, they have arguably failed the competence duty by using a tool without understanding its known and well-publicized failure mode. Competence, in the age of AI, means knowing your tools' weaknesses as well as their strengths, and building your practice around both.

The duty of confidentiality is among a lawyer's most sacred, and the first lesson flagged how AI can threaten it. Here is how to honor it in practice.

The core rule is simple to state: do not disclose client confidential information to a tool that is not appropriately protected. Making that concrete means vetting tools before use and being disciplined about what you enter.

Practical safeguards:

• Vet the tool's data handling. Before using any AI tool with client information, confirm what it does with your input: Is it stored? Used for training? Who can access it? Prefer tools with clear contractual commitments that data is not retained or used to train, and that provide appropriate security.
• Use enterprise or legal-grade tools for confidential work. Consumer versions of general chatbots often lack the necessary protections; enterprise and law-specific versions are typically built to keep your data private.
• Minimize and anonymize where possible. If a general tool is genuinely useful for a task, consider whether you can strip identifying client details first, so no confidence is exposed.
• Consider client communication and consent. Depending on the circumstances and jurisdiction, informing clients about material AI use, or obtaining consent for certain uses, may be appropriate or required.

The underlying principle: the convenience of a tool never outweighs the duty to protect client secrets. A lawyer who would never email privileged documents to a stranger must apply the same caution before pasting them into an AI service whose data practices they have not verified. Confidentiality in the AI era is largely a matter of tool selection and input discipline, deciding in advance which tools are safe for confidential work and never putting client secrets anywhere else.

A lawyer owes a duty of candor to the court and, more broadly, must not present false information. AI's hallucination problem collides directly with this duty, which is why verification is an ethical obligation, not just a quality-control habit.

The rule is absolute: a lawyer is responsible for the accuracy of everything they submit, regardless of whether AI helped produce it. Filing a brief with AI-invented cases is a failure of the lawyer's own duty, and "the AI made it up" is not a defense, courts have made this explicit through sanctions and standing orders requiring lawyers to verify AI output and sometimes to disclose its use.

A useful mental model is supervision. Treat AI output exactly as you would the work of a junior associate: helpful, fast, but requiring your review before it goes out under your name. You would never file a first-year's memo without checking it; the same standard applies to AI, and arguably more so, because AI errs in confident, hard-to-spot ways.

Concretely, this means:

• Verify every citation against the primary source before filing or relying on it.
• Check factual and legal assertions the AI makes, since it can state falsehoods fluently.
• Be aware of court rules on AI use in your jurisdiction, including any disclosure or certification requirements.

The unifying idea is that AI does not dilute the lawyer's responsibility for accuracy; it concentrates it. When a tool can produce convincing falsehoods at speed, the duty to verify becomes more important, not less. The lawyer remains the last line of defense between the AI's plausible output and the court, the client, and the record, and that position cannot be delegated to the machine.

A subtler but important duty concerns reasonable fees and honesty with clients, and AI raises fresh questions here that thoughtful lawyers should address proactively.

The central issue is that AI can make certain tasks dramatically faster. If a task that once took a lawyer five billable hours now takes one hour with AI assistance, billing the client for five hours would be neither honest nor consistent with the duty to charge reasonable fees. The efficiency gain should generally benefit the client, not become hidden margin.

This surfaces several considerations:

• Bill for time actually spent, not time AI saved. Charging for hours that AI eliminated is difficult to justify as a reasonable fee.
• Do not bill clients for learning to use AI tools as if it were work on their matter, absent agreement.
• Be transparent where appropriate about how AI is used, especially if a client asks or if the use is material to the representation. Guidance like the ABA's opinion addresses these fee and communication questions directly.
• Consider passing efficiency to clients, which can be both an ethical stance and a competitive advantage, since AI-enabled efficiency can make legal services more affordable.

The broader point is that AI changes the economics of legal work, and the duties of honesty and reasonable fees require lawyers to handle that change with integrity rather than quietly capturing the savings. A profession built on trust cannot let a productivity tool become a way to overcharge.

Handled well, AI's efficiency is a genuine win-win: the lawyer works faster and can serve more clients, and clients get faster, more affordable service, all consistent with the ethical duties rather than in tension with them.

Turn all of this into something usable: a simple responsible-AI practice any lawyer or firm can adopt, converting the duties into concrete habits.

A workable practice policy comes down to a few durable habits:

• Approve tools deliberately. Decide in advance which AI tools are vetted for confidential work and which are not, and stick to it.
• Verify before you rely. Make checking AI output against real sources an unbreakable step, especially for anything filed or relied upon.
• Keep a human owner. Every piece of AI-assisted work has a lawyer who reviewed it and is responsible for it.
• Stay current. Bar guidance and court rules on AI are evolving; keep abreast of the requirements in your jurisdiction.
• Be honest. With courts about what you file, and with clients about efficiency and fees.

The theme that closes the whole cursus: for lawyers, AI is a powerful assistant governed by the same professional duties as everything else you do. The tools genuinely accelerate research, drafting, and review, but competence, confidentiality, candor, and honesty all continue to bind. The lawyers who thrive with AI are those who treat responsibility as inseparable from capability, using the technology aggressively for leverage while owning every result completely. Get that balance right, and AI becomes a way to serve clients better while fully honoring the duties that define the profession.

Each way AI touches legal work maps to a binding professional duty, competence, confidentiality, candor, and reasonable fees, and a responsible practice turns each duty into a concrete habit with the lawyer always owning the result.