AI agents can fail in too many ways to count. This article by Jerry Lawson focuses on one of the biggest vulnerabilities, prompt injection. However, because there are so many other ways agentic AI can fail, the final sections will also discuss ways to limit the damage a compromised agent or other AI security vulnerability can cause.
Miranda De La Torre, AI and Legal Technology Fellow at ASU’s Sandra Day O’Connor College of Law, discusses the complex challenges of new lawyers now learning powerful systems on the job, often without clear institutional guidance, shared professional norms, or confidence in their own ability to supervise the output.
The majority of law firm employees are using AI with virtually no guidance or guardrails. How does your law firm compare? Do you have an AI policy in place, and have you educated your staff about appropriate AI usage? Nicole Black explains drafting AI governance isn’t as difficult as it might seem, and there’s no better time than now to get started.
Damien Charlotin tracks the claims made by some LegalTech vendors in the past and today with respect to how they handle hallucinations from their offerings. Charlotin is relying on internet-based written marketing material, trying to highlight the changes in how these products are and were presented. The main vendors were a bit more cautious he thought though most still overclaimed in this respect and eventually backtracked, at least implicitly.
Right now the dominant AI strategy in law is using AI to replace or augment human labor on work product. Document review. Contract analysis. Research. First drafts. The logic is straightforward: if AI can do in minutes what an associate does in hours, the firm gets more efficient, margins improve, and clients eventually get lower costs. Every elite firm is running this play. Almost none has reckoned with where it ends. Josh Kubicki proposes an innovative, actionable and success driven deep coverage alternative that re-frames the institutional infrastructure around both the partners and the clients.
Despite its unpopularity and the availability of other citation manuals, The Bluebook remains widely used at many law schools to teach legal citation format to law students, and it is relied on by law reviews and courts. The twenty-second edition of The Bluebook was released in May 2025. This new edition includes a new rule—Rule 18.3—that crafts a citation format for legal writers to use when citing generative artificial intelligence (“AI”). This Book Review by Prof. Jessica R. Gunder proceeds in three parts. First, it examines the purpose of citations in legal writing and identifies circumstances in which the citation of generative AI output is appropriate. Second, it considers what The Bluebook requires of authors using generative AI technology and why The Bluebook’s requirements are inappropriate, focusing on: (1) errors within Rule 18.3 itself; (2) the unreasonable burden Rule 18.3 imposes; (3) Rule 18.3’s incompatibility with how generative AI technology is actually used; and (4) how the requirements imposed by Rule 18.3 violate attorney-client confidentiality requirements and work product protections. Third, and finally, it discusses why The Bluebook’s flawed approach matters and how it might be addressed.
Michael Berman addresses benchmarks used for AI legal research platforms in the context of the risk of hallucinations in retrieval-augmented generation (RAG) AI outputs. As Berman states, verification, of course, is not only good advice, but also an ethical mandate.
A lawyer’s ability to stand behind their legal work is a real advantage over legal AI. But for many clients, paying more to transfer risk to a lawyer is a luxury — and maybe soon, an unnecessary one. Jordan Furlong’s opening keynote at ABA TECHSHOW March 26, 2026 in Chicago addressed two critical questions facing the legal profession right now: “As AI displaces lawyers from a growing share of legal task performance, what will be left for lawyers to do?” and “How are we going to develop lawyers when we don’t know what we’re training them for, and when we can’t count on law firms to do the training anymore?”
Josh Kubicki identifies a significant risk that will impact large law firm services to global enterprise wide clients. As Kubicki details, the single largest buyer of elite legal services in the world is now funding the construction of an AI law firm designed to do that same work. This not hype. This is not a pilot program. Blackstone and Norm AI are, per their own public announcement, “collaborating to shape and develop Norm Law legal services for Blackstone’s use.” The client is co-designing the firm that will compete with its own outside counsel. And it’s not being subtle about it.
Knowing the difference between a general AI tool and one trained on specific sources can mean the difference between getting an accurate answer and becoming quickly frustrated with outcomes that either don’t answer the question thoroughly or answer the question in a confused mixture of fact and fiction. While not always clear, the data that lies behind the GenAI tool is just as important to consider as the user interface or the cost. Without trustworthy or relevant underlying information, the resulting AI-generated output will be less helpful or less trusted and result in inefficiencies as lawyers and staff work to fill the gaps in the GenAI’s response. Considering these factors, Kristopher Turner’s article identifies how and why a retrieval augmented generation (RAG) can give focused and highly specific answers related to one area that someone needs to quickly understand.