How is AI changing the skills expected of professionals in your jurisdiction?
AI is already changing the skills expected of professionals, including attorneys. I believe it is making lawyers more efficient by providing useful roadmaps to statutes, precedent and transactional templates. It is also making it easier to quickly complete more mundane tasks (such as creating a table of contents or authorities).
My favourite aspect of AI is its ability to efficiently condense voluminous amounts of information into granular detail. For example, in a complex legal case, AI can assist in generating timelines from thousands of pages of documents. In an M&A transaction, it can help create a workable diligence or closing checklist. These tasks can save counsel invaluable amounts of time and resources.
However, the professional skill required is not knowing how to use AI. It is knowing how to use it well. AI may provide a starting point, but attorneys must still verify sources, review outputs, assess context and determine whether the answer provided is legally and factually sound. In that sense, AI is not reducing the need for professional expertise. It is, however, increasing the need for disciplined review.
When professionals understand the technology’s limitations and use it as a tool to enhance independent judgement, it is a massive timesaver. The issues arise when it is trusted blindly.
In California and Nevada, attorneys remain bound by their professional obligations regardless of the technology used. AI can assist with research and organisation, but it does not alter the lawyer’s duties of competence and confidentiality.
What governance or supervision structures are firms and businesses putting in place to manage AI responsibly?
Firstly, firms should carefully ensure that prompts provided to any large language model (LLM) are not retained or shared with third parties – including being retained by the platform for recursive learning, model training or self-improvement. This is especially important where prompts may contain privileged, confidential, proprietary or otherwise sensitive information.
Second, we recommend that firms and businesses pick a platform, thoroughly vet it and then stick with it consistently throughout the firm or business. Permitting different employees to use different AI tools without oversight creates unnecessary risk. One employee may use a platform with strong privacy protections, while another may use a public tool that retains prompts or uses inputs for training. From a governance standpoint, that inconsistency is dangerous.
Vetting should include a careful review of the provider’s terms of use, privacy policy, data-retention practices, security protocols and confidentiality commitments. Firms should understand where data is stored, whether it may be reviewed by humans, whether it may be used to improve the model and what contractual remedies exist if the platform mishandles information.
Finally, there should be written AI-use policies in place to guide employees on responsible AI use. These policies need to address which platforms may be used, what information may or may not be uploaded, when human review is required and who is responsible for approving AI-assisted work before it is circulated externally. A policy that simply sits in a file and is never explained is not enough. Employees should be trained on it, and the policy should be updated in practice as the technology develops.
All of this sits against California’s developing AI and privacy framework. The California Consumer Privacy Act already places significant emphasis on data handling, while newer AI legislation (including SB 53) reflects California’s focus on transparency, risk management and accountability in AI use. Here, having these internal policies is especially important.
What liability or risk issues arise when professionals rely on AI-generated analysis or advice?
The biggest risks involve breaches of professional obligations, including duties of confidentiality and attorney-client privilege. If a lawyer or client uploads privileged communications into a public AI platform, those protections may be lost or severely weakened. Even if privilege is not automatically waived in every circumstance, the risk is substantial enough that professionals should proceed with extreme caution.
There is also significant work-product risk if the results of a prompt are not thoroughly reviewed and vetted for accuracy. LLMs have become notorious for producing responses that contain errors, unsupported assumptions and hallucinations. In the legal context, that can be especially dangerous. A fabricated case citation, incorrect statutory summary or inaccurate description of controlling law can create professional embarrassment, sanctions exposure and client harm.
But the risk is not limited to obvious hallucinations. Sometimes the greater danger is context. AI may generate a response that is partially correct but incomplete, or accurate in one jurisdiction but wrong in another. It may miss a factual nuance, overlook an exception or apply the wrong legal standard.
For that reason, professionals should never treat AI-generated analysis as final advice. It should be treated as nothing other than a starting point in this context. Attorneys remain responsible for their work product, their advice and their filings. The fact that a machine generated the first draft does not transfer responsibility away from the professional.
I am also hesitant to see professionals over-rely on AI because doing so may influence research creativity and advocacy. Legal analysis is not always about retrieving information. Judgement, strategy, interpretation and persuasion are all skills which legal professionals must continue to hone. AI is powerful and useful, but it must assist professional work, not replace professional responsibility.
