Is it ethical to use AI on client files? ABA Opinion 512, explained

In July 2024 the American Bar Association issued its first formal ethics opinion on generative AI. If you're a solo attorney wondering whether you can safely use AI on real client files, this is the document that matters — and the short version is more permissive than many lawyers fear.

What Opinion 512 actually says

Opinion 512 maps generative AI onto the existing Model Rules of Professional Conduct. It doesn't create new rules; it applies old ones. The duties it emphasizes are:

• Competence (Rule 1.1). You must understand, at a reasonable level, the benefits and risks of the AI tool you use — including where your inputs go and how they're handled.
• Confidentiality (Rule 1.6). This is the heart of it. Before you enter information relating to a client's representation into an AI tool, you must consider whether doing so risks disclosure to people outside the representation.
• Communication and consent (Rules 1.4 / 1.6). In many cases you need the client's informed consent before feeding their information into a self-learning or third-party AI system.
• Supervision (Rules 5.1 / 5.3). AI output is a draft, not a work product you can rely on blindly. You're responsible for verifying it.
• Reasonable fees (Rule 1.5). If AI saves you hours, you generally can't bill the client as though it didn't.

Why confidentiality is the real hurdle

Most of the practical anxiety around legal AI comes down to one question: where do my client's documents go? With a typical cloud AI tool, your inputs travel to a vendor's servers. Depending on the terms of service, that vendor may retain the data, have employees who can access it, or use it to train future models. Opinion 512 flags exactly this: you have to read the terms, understand the data handling, and weigh whether using the tool risks a confidentiality breach.

That's a real burden. Terms of service change. "We don't train on your data" can quietly become "we may use anonymized data to improve our services." For a solo handling sensitive matters — medical records, financial disclosures, family-law filings — that ongoing diligence is exactly the kind of thing that's easy to get wrong.

How on-device tools change the analysis

Here's the key insight: most of Opinion 512's confidentiality and consent analysis is triggered by disclosure of client information to a third party. If the AI runs entirely on your own computer — no upload, no vendor server, no external model — there's no third-party disclosure to analyze. The data never leaves your control.

That doesn't make you exempt from the opinion. You still owe competence (understand the tool) and supervision (verify the output). But it removes the single thorniest issue: you're not handing a client's confidential file to anyone, so the consent-before-disclosure question largely falls away.

A practical checklist for solos

1. Know where the data goes. For any AI tool, answer one question first: does my client's document leave my computer? If yes, read the terms and consider consent. If no, you've removed the biggest risk.
2. Verify everything. Treat AI output as a first draft from a sharp but unaccountable assistant. Check the dates, the citations, the summaries.
3. Document your diligence. Keep a short note on which tools you use and why you concluded they're consistent with your duties.
4. Check your state. The ABA opinion is influential but not binding; several states have issued their own guidance. Follow your jurisdiction.

The takeaway for a modern solo practice: AI is usable today, and the cleanest path through the ethics rules is to keep client data on your own machine. That's the entire design premise behind on-device tools like ClerkSafe — the confidentiality question answers itself when nothing is ever uploaded.

This article is general information, not legal advice. Read ABA Formal Opinion 512 and your state's rules and guidance for the authoritative requirements.