7 Things Your Firm Needs Before Using AI
A compliance checklist for small law firms
79% of legal professionals use AI. Only 21% of firms have an AI policy.
If your staff is experimenting with ChatGPT and you don’t have these seven things in place, you have a compliance gap. Not a theoretical one. A real one, with malpractice exposure, billing ethics risk, and client confidentiality on the line.
This guide gives you the checklist. Seven items, each with a clear standard for what “good” looks like. Some are backed by specific California rules of professional conduct. Others are best practices that protect you if something goes wrong. We label each one so you know the difference.
No product pitches, no scare tactics. Just the framework your firm needs before anyone touches an AI tool with client data.
Ben McEachen is the principal of Henry Street Advisors, where he helps small and mid-market firms redesign their operations for the AI era. He brings 28 years of experience building and scaling businesses, with practitioner-level AI fluency, not just theory.
The 7-Item Compliance Checklist
1. Written AI Use Policy
Supports: Rule 5.1 (supervisory duties), State Bar GenAI Practical Guidance
Without a written policy, 79% of your team is making their own rules. That’s not a technology problem. That’s a governance problem.
Your AI use policy should define which tools are approved for use (and which are explicitly prohibited), what types of data can and cannot be processed through AI systems, who has authority to approve new AI tools, and what happens when someone violates the policy. It doesn’t need to be 20 pages. A two-page document that staff actually read is better than a binder nobody opens.
The California State Bar’s GenAI Practical Guidance (November 2023) calls on supervisory lawyers to “establish clear firm policies on permissible AI uses and provide training on ethical pitfalls.” This is advisory guidance, not a binding rule, but it reflects how the Bar interprets your existing supervisory duties. If you don’t have a written policy, you’re behind what the Bar expects.
2. Client Informed Consent Process
Required by: Rule 1.4 (communication), reinforced by State Bar GenAI Practical Guidance
Rule 1.4 requires lawyers to keep clients reasonably informed about significant developments in their matter. The State Bar’s GenAI guidance extends this to AI, calling on lawyers to “communicate transparently with clients about use of AI in their matter, costs associated with AI use, limitations of AI tools, and risks.” The guidance is advisory, but the underlying duty of communication is binding.
A solid informed consent process includes: a clear explanation of what AI does in the client’s matter (and what it doesn’t do), disclosure of any third-party vendors who access case data, a description of the security measures protecting their information, acknowledgment of AI limitations (including the risk of inaccurate output), and the client’s right to decline or withdraw consent at any time.
This isn’t just a compliance checkbox. “We use AI for research and first drafts, which lets us move faster and keep your costs down. Here’s how we supervise the output.” That’s a client development conversation, not a liability disclosure.
3. Vendor Confidentiality Agreement
Supports: Rule 1.6 (confidentiality of information)
Rule 1.6 requires you to protect client confidences. When you use a third-party AI tool, you’re sharing client data with a vendor, and the State Bar guidance says you must ensure that vendor either (a) won’t share it with third parties, (b) won’t use it for training, or (c) that you’ve anonymized the data first. A vendor agreement is how you document that you’ve met this duty.
But a “no training on your data” clause alone is not enough. Other risks remain even with that clause in place:
- Data-in-transit exposure. Data travels across networks and sits on provider servers during processing. A provider-level breach exposes whatever you sent, regardless of training agreements.
- Employee access. Provider employees may access data for debugging, content moderation, or security reviews. “No training” doesn’t restrict human access.
- Breach risk. Providers can be breached. The no-training clause doesn’t prevent attackers from exfiltrating data from provider infrastructure.
- Subpoena and discovery. Everything sent to an AI provider could be discoverable in litigation.
- Malpractice defense. If a breach occurs and you relied solely on contractual protections without technical controls (like data anonymization), liability is harder to defend.
Your vendor agreement should include: an approved vendor list with contractual guarantees from each provider, an anonymization protocol for sensitive data, audit rights, a breach notification timeline, data retention and destruction policies, and a prohibition on using client data for model training or improvement. (Note: California’s SB 446, effective January 1, 2026, shortened the state’s breach notification deadline to 30 days. Your vendor agreement should reflect this timeline.)
4. Active Attorney Oversight
Required by: Rule 5.3 (supervision of nonlawyers)
This one is binding. California Rules of Professional Conduct Rule 5.3 requires a lawyer with supervisory authority over a nonlawyer to “make reasonable efforts to ensure that the person’s conduct is compatible with the professional obligations of the lawyer.” The State Bar’s GenAI guidance extends this supervisory duty to staff using AI tools and to third-party vendors processing client data.
Passive contracts don’t satisfy this requirement. Active oversight means: establishing review protocols for all AI-generated work product before it’s used, documenting who reviewed the output and what changes were made, setting clear boundaries on what staff can and cannot do with AI tools, and regularly checking whether policies are being followed.
The attorney can’t outsource ethical responsibility to a tool or a vendor. Every AI-assisted output needs a human review step before it touches a client matter, a court filing, or a billing entry.
5. Document Your AI-Assisted Work
Best practice: supports Rule 5.3 compliance and breach response obligations
No California rule or State Bar guidance explicitly mandates an “audit trail system” for AI use. But two things make documentation a smart defensive habit.
First, Rule 5.3 requires you to supervise AI-assisted work. If a client or the Bar ever questions whether you actually reviewed AI output before using it, your documentation is your evidence. A note in the case file is infinitely better than “I’m sure I looked at it.”
Second, if a data breach ever occurs, California’s SB 446 (effective January 1, 2026) requires you to notify affected individuals within 30 days. Knowing what client data you sent to which AI tool, and when, is the difference between a targeted notification and a panicked scramble through months of history.
This doesn’t need to be a formal logging system. For a solo practitioner, it can be as simple as: a note in your case management system when AI was used on a matter, what tool you used, and confirmation that you reviewed the output. Build the habit now while volume is low.
6. Billing Transparency Protocol
Required by: Rule 1.5 (fees), State Bar GenAI Practical Guidance
This is the one most practitioners feel in their gut. The State Bar guidance is explicit: lawyers “may charge for time actually spent on AI-assisted work (refining prompts, reviewing outputs) but must NOT charge for time saved by AI.”
If a first-draft motion used to take four hours of paralegal time and AI generates it in 20 minutes, you can’t bill four hours. You can bill for the attorney review, prompt refinement, and substantive revision that turned that draft into final work product. The client pays for your judgment, not for a machine’s output.
Here’s the reframe. This isn’t a revenue cut. It’s a pricing redesign opportunity. Flat-fee billing lets you capture the value of your efficiency gains. Fifty-nine percent of firms now bill flat fees exclusively or alongside hourly, according to Clio’s 2024 data. Firms billing flat fee collect payment twice as fast and close matters 2.6 times faster. And 71% of clients say they’d prefer a flat fee over hourly billing.
The firm that prices a flat-fee motion at $800 when the old hourly cost was $1,200, and produces it in a quarter of the time, just improved margins and won the client.
7. Staff Training Program
Best practice: supports Rule 5.1 and Rule 5.3 compliance
Who was trained, on what, when. No rule mandates a formal training program, but this is what protects the firm if a staff member misuses AI. The State Bar guidance calls on supervisory lawyers to “provide training on ethical pitfalls.” If a paralegal pastes client data into consumer ChatGPT and you never trained them not to, that’s on you under Rule 5.3.
Your training program should cover: which tools are approved and how to use them, what data never touches consumer AI (free-tier ChatGPT, Gemini, etc.), how to verify AI output before using it in any client matter, and what to do if they’re unsure whether a use case is permitted.
Document every training session: date, attendees, topics covered, materials provided. This turns compliance from a burden into a competitive advantage. When a prospective client asks “how do you handle AI?”, you can show them a documented program instead of a vague reassurance.
The Cost Reality
The expensive part isn’t the tools. It’s the implementation.
| Approach | Annual Cost (Per Attorney) | What You Get |
|---|---|---|
| Specialized legal AI products (Paxton + CounselPro + practice management upgrade) | $4,500 - $8,100 | Point solutions for drafting, financial discovery, and case management. No integration between them. According to Axiom Law’s analysis, 95% of legal AI pilots fail to deliver measurable impact. |
| Custom-configured general-purpose stack (Claude Team + workflow automation + existing practice management) | $600 - $1,800 | AI configured for your specific practice, integrated with your existing tools, with governance and compliance built in. |
The specialized products aren’t bad. They’re just expensive, and they don’t solve the implementation problem. A small firm doesn’t need five subscriptions. It needs someone to design how AI fits into the way the firm actually works, then build it from tools the firm can afford.
Red Flags: If Any of These Are True, Act Now
- Staff using consumer ChatGPT (free tier) with client data
- No written AI policy in place
- Billing model unchanged despite AI adoption
- Vendor contracts that say “no training” but don’t address the other risks listed above
- Clients haven’t been informed about AI use in their matters
- No documented staff training on AI tools and policies
- AI-generated output going into filings without attorney review
If three or more of these apply to your firm, you have a gap that’s growing every month your staff experiments without guardrails.
What’s Next
This checklist is the starting point. The seven items above give you a framework, but the hard part is implementation: writing the policies, selecting the right tools, configuring them for your practice, training your team, and building the habits that make compliance stick.
Have a question about any of this? What’s the thing you’re most unsure about? I read every one. Get in touch.
Want the full picture? Read “AI Just Compressed Your Production Costs. Now What?” for the deeper analysis of what AI means for legal billing, pricing, and competitive positioning.
Ben McEachen is the principal of Henry Street Advisors, a consultancy that helps small and mid-market companies redesign their operations for the AI era. With 28 years of experience building and scaling businesses, Ben combines deep operational expertise with practitioner-level AI fluency to help firms adopt AI the right way: compliant, practical, and built to last. He is not a lawyer. This guide is not legal advice.