Artificial intelligence is reshaping law offices at breakneck speed, but the gap between technical efficiency and professional liability has never been narrower. At the center of this transition is Pearl, the hybrid intelligence powerhouse connecting advanced AI infrastructure with a network of over 20,000 verified licensed professionals across 100+ disciplines.
By pairing automated workflows with human-in-the-loop subject matter experts, Pearl processes millions of complex consultations while powering specialized practice tools like Fount, an intake, billing, and scheduling platform designed specifically for solo practitioners and boutique law offices.
To discuss the boundaries of AI delegation, regulatory fragmentation, and the myth of the “AI exception,”
LawFuel sat down with Nick Tiger, (left) Deputy General Counsel at Pearl.
1. How should lawyers validate generative AI outputs?
“AI can generate the work. It can’t outsource the lawyer’s judgment. The more consequential the work, the more important it is that the right lawyer can independently stand behind it.”
Nick Tiger: Whether you have already begun implementing AI into workflows or have no idea where to start and are still experimenting with the technology, lawyers should always treat AI like a first-year associate whose work should be checked and verified before you attach your name to it. AI can help with research, summarization, administrative tasks, and early drafts, but this work should always be human-verified, and AI should never substitute for the lawyer’s strategic or final professional judgment.
Lawyers validating GenAI outputs should go beyond baseline proofreading: it requires reviewing the output for obvious errors or omissions, but also checking sources, citations, and the underlying context provided to the AI in the prompt.
As the consequences of client impact rise, the validation should become more substantive. For example, an output that contains analysis that could shape client advice, litigation strategy, or a court filing requires deeper validation than an output that provides a routine, lower-consequence research summary.
Possessing a law license does not mean the lawyer is automatically qualified to validate AI outputs. Meaningful validation requires that the lawyer have the task-relevant expertise, information, and context necessary to identify a material error.
A corporate lawyer cannot meaningfully validate specialized tax advice merely because both involve law. Depending on the matter, that lawyer may need to gain competency in the area prior to validation or escalate to another lawyer with deeper expertise.
2. What changes when AI becomes agentic?
“With generative AI, you’re often validating an answer. With agentic AI, you also have to validate the authority you’re giving the system.”
Nick Tiger: With generative AI, there is a flashpoint where a lawyer can review what the AI is saying before deciding to rely on it. Agentic AI changes the risk because the system can move from saying to doing. It may access client information, communicate externally, use tools, or initiate actions before a lawyer reviews the result.
Agentic AI supervision has to address two different questions:
- Was the output right?
- Should the AI have been authorized to take that action in the first place?
The framework should remain risk-based. Let AI move quickly on lower-consequence, reversible tasks—especially when acting in an administrative capacity.
As agentic AI actions become more substantive, harder to reverse, or capable of materially affecting legal rights, money, confidentiality, or client interests, the lawyer should increase the number of human checkpoints and narrow the system’s independent authority.
The lawyer reviewing the agentic AI system and its outputs must have the relevant information, expertise, and authority to assess what the agent proposes to do, approve it, or change, stop, and escalate it.
3. What do lawyers’ existing professional obligations already require?
“AI may be new. A lawyer’s accountability isn’t. There is no AI exception to the rules that already govern lawyers.”
Nick Tiger: The useful starting point is not to invent a separate ethics code for AI. The ABA Model Rules of Professional Conduct already impose clear standards:
- Rule 1.1: Competence
- Rule 1.4: Client Communication
- Rule 1.5: Fees
- Rule 1.6: Confidentiality
- Rule 3.1: Meritorious Claims
- Rule 3.3: Candor toward the Tribunal
- Rules 5.1 & 5.3: Supervision of Partners, Associates, and Non-Lawyer Assistance
ABA Formal Opinion 512 applies these longstanding Model Rule duties directly to generative AI technologies. The critical takeaway in Opinion 512 is that AI doesn’t displace professional judgment—it makes professional judgment about how to use and validate the technology part of competence itself. AI can do more of the work, but it cannot inherit the lawyer’s professional responsibility for that work.
The technology changes who produces the first draft; it doesn’t change who owns the legal judgment.
While binding requirements depend on local jurisdictions, the baseline principle is clear: lawyers must understand the tools well enough to use them competently, protect client data, review work rigorously, and stand behind what they submit or advise.
4. What emerging AI requirements should lawyers be watching?
“For the foreseeable future, there probably won’t be one AI rule for every lawyer. Firms need a lightweight regulatory radar that tracks professional rules, courts, and legislation separately.”
Nick Tiger: Lawyers must monitor AI developments from three distinct sources that rarely move in lockstep: regulators, courts, and legislatures. A rule established by one judge or state bar should never be assumed to apply across the board.
| Regulatory Source | Primary Focus | Key Consideration |
| Professional Regulators | State Bar Ethics & ABA Opinions | Duties under Competence, Confidentiality, and Supervision |
| Courts & Judges | Local Standing Orders & Practice Rules | Mandatory disclosures, independent verification of citations |
| State Legislatures | Statutory Compliance & Liability | Verification mandates and nondelegation of practice rules |
Tracking this does not require a dedicated regulatory division. Assign one team member to maintain a simple matrix documenting:
- Jurisdiction and governing authority
- Effective date
- Affected AI activity
- Core requirement and corresponding firm policy update
Consider two concrete examples:
- California SB 574: If enacted, it would impose express statutory mandates concerning verification of GenAI output, citation checking, client confidentiality, court submissions, and nondelegation of legal practice.
- New York Part 161 (Effective June 1, 2026): Establishes a statewide framework under which courts can adopt a model local AI rule. New York does not treat AI use as universally requiring disclosure, but lawyers practicing before adopting courts must independently guarantee that citations and assertions are authentic.
5. How does Pearl approach human validation?
“AI should scale. Human judgment should scale with the consequences.”
Nick Tiger: Pearl’s platform allows users to triage complex issues with AI and seamlessly connect with an expert network spanning 100+ specialties, including legal practice. When AI meets human expert oversight, answers are rated 22% more helpful than standalone models like ChatGPT.
This experience led Pearl to implement a two-part operational test:
- What happens if the AI is wrong?
- Can the person relying on the output independently identify the error?
When the stakes are low, AI access should be broad and aggressive—serving as an unfiltered force multiplier for speed. But when AI outputs deliver consequential, individualized guidance that the recipient cannot evaluate on their own, qualified human judgment must step into the loop before a hallucination causes material reliance or harm.
6. Could California’s approach become the de facto model for lawyers elsewhere?
“California can influence the national conversation without becoming the national rule.”
Nick Tiger: California SB 574 passed both chambers of the Legislature and went to Governor Gavin Newsom, but has not yet become law.
Firms should follow its progress without assuming every California mandate will—or should—become the national standard. The question to ask is whether particular concepts start recurring across multiple states, courts, and bar associations. A durable standard does not emerge because one state moved first; it emerges because the underlying principle repeatedly proves its utility across multiple jurisdictions.
Watch SB 574 for its granular handling of citations, verification, disclosure, and confidential data, rather than treating it as an automatic blueprint for other states.
7. What are lawyers most likely to get wrong about client confidentiality when using AI?
“Security is a product feature. Confidentiality is a professional obligation. Lawyers need to understand both.”
Nick Tiger: The most common mistake is reducing confidentiality to a binary: either believing AI must be banned outright, or assuming a vendor’s “enterprise” badge solves every problem. Neither reflects competent professional analysis.
Under ABA Model Rule 1.6 and Formal Opinion 512, lawyers must make reasonable efforts to prevent unauthorized disclosure of, or access to, information relating to representation—including former and prospective clients.
Lawyers must evaluate what actually happens to data fed into a tool:
- Is the data retained on third-party servers?
- Can vendor staff or outside contractors view prompt histories?
- Is input data ingested to train or refine public or proprietary models?
- What specific contractual, encryption, and tenancy controls govern the workspace?
- Who owns and controls data deletion protocols?
Opinion 512 does not hold that every AI prompt demands informed client consent. That analysis depends on the tool, the data sensitivity, and the context. However, where representation-related data creates material confidentiality exposure, informed consent may be required—and generic boilerplate buried in an initial engagement letter will rarely satisfy that duty.
8. What should a solo lawyer or small firm put in place right now?
“A small firm doesn’t need an enterprise AI department. It needs a few decisions made in advance.”
Nick Tiger: The most effective policy for a small firm isn’t the longest; it’s the one a busy practitioner can follow under pressure. Decide four fundamentals before you face a filing deadline:
- Approved Toolset: Which specific AI tools are authorized for firm work?
- Data Boundaries: What categories of firm and client data are strictly barred from AI prompts?
- Checkpoints: Where in the workflow must a human review happen?
- Final Ownership: Who signs off on the final deliverable?
Keep your enforcement proportional. Do not strangle productivity by reviewing a routine procedural summary with the intensity of a dispositive motion.
When implemented sensibly, AI helps solo and boutique practitioners offload administrative overhead. Practice-tailored tools like Fount (powered by Pearl) automate client intake, scheduling, follow-ups, and billing.
Handing off mechanical friction allows solo attorneys to run lean, close the administrative gap with larger firms, and focus their time on what clients actually pay for: strategic legal judgment.
