AI safety for lawyers: It's not how the engine works, it's how you drive the car

When a lawyer drives to court, nobody expects them to understand how fuel injection works. Nobody cares if they can explain the mechanics of anti-lock brakes or the physics of combustion engines. But everyone absolutely expects them to check their mirrors, wear a seatbelt, and drive defensively.

The same principle applies to artificial intelligence in legal practice.

The mechanics don't matter (much)

Many articles about AI in law focus on the fascinating technical details -- how large language models map words in thousand-dimensional spaces, how they use pattern recognition to predict text and how they're essentially sophisticated autocomplete systems. Interesting? Yes. Relevant to daily practice? Rarely.

Consider this: multiple lawyers have been sanctioned for submitting AI-generated fake cases to courts. Last month, an attorney used ChatGPT to "enhance" an already-written brief, but never read the final version, missing the fabricated cases ChatGPT had added. None of these sanctions occurred because the lawyer failed to understand language models or neural networks. They happened because lawyers violated basic operational safety rules.

Just as driving safety isn't about understanding engineering, AI safety in law isn't about understanding algorithms. It's about following three simple non-negotiable protocols that prevent catastrophic mistakes.

1. Check your mirrors: Verify every citation

Driving rule: Check your mirrors before changing lanes. You need to know what's actually around you, not what you assume is there.

AI equivalent: Verify every legal citation AI provides. Every case, statute and quote must be independently confirmed.

AI doesn't lie maliciously -- it generates plausible-sounding content based on patterns. Just as we check mirrors to see what's around us, verification reveals what AI actually found.

This isn't paranoia; it's professional responsibility. Would you change lanes just because your passenger said it looked clear? Then don't submit a brief without checking citations just because AI said they exist.

2. Wear your seatbelt: Write careful prompts

Driving rule: Always wear a seatbelt. It protects you when things go wrong, which they inevitably will.

AI equivalent: Write prompts that actively guard against bias and sycophancy.

AI systems are designed to be helpful and agreeable. Ask them "Is my argument strong?" and they'll often say yes, regardless of your argument's actual merit. Ask them to find supporting cases for a weak position, and they'll try their best to comply -- even if it means fabricating citations.

Just as seatbelts protect against collisions, careful prompts protect against AI's eagerness to please. Instead of asking "How can I win this motion?" ask "What are the three strongest counterarguments to this position?" Instead of "Find cases supporting my theory," try "What would opposing counsel argue?"

Good prompts assume AI will try to please you, so get it to challenge your work, not validate it.

3. Drive defensively: Treat AI like a first-year associate

Driving rule: Drive defensively. Expect mistakes from others and plan accordingly.

AI equivalent: Treat AI output like a first draft from a smart but inexperienced first-year associate or extern.

A new lawyer is smart, well-educated and eager to help. They might produce impressive work quickly. But they lack judgment, make confident assertions about things they don't fully understand, and may confuse legal-sounding language with legal reasoning.

You wouldn't file that new lawyer's first draft without review and revision. You'd revise, question and test their work. Do the same with AI. Use it to accelerate drafting or generate ideas but apply the same editorial scrutiny you'd use with any junior colleague's work. Ask: Is this reasoning sound? Are these citations accurate? Would I stake my reputation on this?

The real danger isn't technical ignorance

The lawyers facing sanctions didn't fail because they couldn't explain how neural networks function. They failed because they treated AI like a trusted senior colleague, not an overeager assistant. These attorneys essentially handed AI the keys and never checked where it drove -- discovering too late that their "enhanced" briefs had veered into fabrication.

These aren't technical failures requiring deep expertise to solve. They're professional discipline failures.

This is professional responsibility, not technology law

These aren't technology rules -- they're California Rules of Professional Conduct requirements that happen to involve technology. The State Bar of California's November 2023 AI guidance makes clear that existing professional responsibility obligations apply fully to AI use, including Rules 1.1 (competence), 1.3 (diligence), 1.6 (confidentiality), and 3.3 (candor to tribunals).

Rule 1.1 specifically requires lawyers to "keep abreast of the changes in the law and its practice, including the benefits and risks associated with relevant technology." When you submit a brief with fabricated citations, State Bar disciplinary authorities won't ask whether you understand neural networks. They'll ask whether you exercised the competence and diligence required under Rules 1.1 and 1.3.

California's official guidance is explicit: "A lawyer must critically review, validate and correct both the input and the output of generative AI" and "A lawyer's professional judgment cannot be delegated to generative AI." Technology doesn't change these fundamental obligations. The same rules apply to any tool lawyers use.

The State Bar emphasizes that lawyers must understand AI's limitations and avoid overreliance on these tools while maintaining their professional judgment. This isn't about becoming a tech expert. It's about applying the same professional standards you'd use with any research tool that could impact your clients' interests.

The Rules of Professional Conduct already provide the framework. The challenge isn't learning new rules -- it's following existing ones when using AI.

Simple rules for complex technology

The beauty of the driving metaphor is its universality. Whether you're driving a 1995 Honda Civic or a 2025 Tesla with full self-driving capability, the basic safety rules remain the same: check your surroundings, protect yourself against foreseeable problems and stay alert for the unexpected.

Similarly, whether you're using ChatGPT, Claude, or whatever AI system comes next, the operational safety rules remain constant:

1. Always verify citations (check your mirrors)

2. Write prompts that resist bias and sycophancy (wear your seatbelt)

3. Treat all AI output as a first draft requiring human judgment (drive defensively)

None of these require knowing how AI works. They just require knowing how to work safely with it.

Focus on what actually matters

When someone wants to explain AI safety through technical architecture, ask this: How many sanctioned lawyers would have been saved by understanding large language models or neural networks?

The answer: zero.

How many would have been saved by checking their citations?

Every single one.

AI safety in law isn't about the engine under the hood. It's about how we handle the wheel. Competent, ethical driving is what keeps us -- and our clients -- safe.

Disclaimer: The views expressed in this article are solely those of the author in their personal capacity and do not reflect the official position of the California Court of Appeal, Second District, or the Judicial Branch of California. This article is intended to contribute to scholarly dialogue and does not represent judicial policy or administrative guidance.