AI ETHICS The vendor evaluation process differs between established companies and emerging startups. Established providers like Microsoft, Google, and Thomson Reuters bring reputational stakes and institutional resources to security and compliance. However, their products may be more generic and their terms less negotiable. AI startups may offer more tailored solutions and greater flexibility but carry higher risk of business failure, security immaturity, or inadequate resources to honor contractual commitments. Defenders should weigh these tradeoffs carefully and consider requiring startups to demonstrate security certifications, insurance coverage, or escrow arrangements protecting against business discontinuity.
3.
Low-Risk vs. High-Risk Tasks
Not all AI applications carry equal risk. Defenders should calibrate their AI use based on the sensitivity of the task and the consequences of error.
Low-risk applications involve tasks where AI errors are unlikely to harm clients and where confidential information need not be disclosed to the AI system. Examples include summarizing general legal principles not specific to any case, drafting administrative documents like office policies or internal memoranda, generating templates for routine correspondence, creating educational materials for staff training, and brainstorming approaches to common legal issues. These tasks can often be performed with consumer-grade AI tools, though enterprise systems remain preferable where available. Human review of the output can eliminate the risk associated. Medium-risk applications involve work product that affects client matters but can be thoroughly verified before use. Drafting initial versions of motions or briefs falls into this category. AI can provide a starting point that the defender then reviews, revises, and validates. Research assistance, such as identifying potentially relevant authorities or summarizing long documents, similarly offers efficiency gains while preserving human oversight. The Vanderbilt Project on Prosecution Policy framework identifies “litigation content” as one category where AI has shown promise for prosecutors, including to draft jury selection questions, opening statements, direct examination questions, motions, and appellate arguments. 70 Defenders engaging in medium-risk applications must verify AI outputs rigorously, as discussed below, and should prefer enterprise systems with appropriate confidentiality protections. High-risk applications include, but are not limited to, tasks involving the use of privileged and/or confidential information and where AI errors could directly harm clients or where the work requires judgment, strategy, or analysis that AI cannot reliably provide. Case-specific factual analysis, strategic decision-making, client counseling, and courtroom advocacy fall into this category. AI should support rather than supplant human judgment in these areas.
36
Parity in Practice: The Defender’s Duty to Ethically Use AI
Made with FlippingBook Online document maker