According to a document from December 2025, the American Bar Association’s Task Force on Law and Artificial Intelligence has published its Year 2 Report on the Impact of AI on the Practice of Law, and the document reads less like a victory lap and more like a carefully managed moment of transition. The report closes the Task Force’s two-year run while signalling that the work “will continue in the Center for Innovation,” which is a polite way of saying that the profession cannot treat AI as a temporary disruption that will eventually calm down.
The profession moved on from “should we use it” to “how do we survive using it”
The report describes a visible shift in professional psychology. The report notes that last year’s debate centred on ethics and confidentiality, while this year’s debate has moved to operational questions about how to use AI, and how to do so without becoming worse lawyers with better software. The report frames that shift as positive, because lawyers now focus less on replacement anxiety and more on using AI to become “better lawyers.”
The report also refuses the comforting idea that “easy to access” means “easy to adopt.” The report emphasises that firms face a genuine change-management problem, because rapid user adoption still leaves a persistent need for support, training, and workflow design. The report uses a memorable internal diagnosis, because a segment of lawyers treat “self-service” as “no service,” and that cultural reality becomes a structural constraint on AI deployment inside organisations.
The next competitive divide is not just technology, it is governance plus talent
The report predicts stratification into technology “haves” and “have-nots.” The report links that divide to a practical dilemma: consumer-grade tools become more capable, but they create privacy and confidentiality risks, while enterprise tools can be hosted in more secure environments but require expensive licensing and additional spending on skilled staff. The report also notes an emerging “arms race” for scarce human expertise, which is a revealing phrase in a sector that still likes to imagine itself as primarily about judgement rather than infrastructure.
The report adds a second, less glamorous constraint. The report describes how the promise of feeding a tool “the entire body of a firm’s work product” collides with reality, because document repositories contain obsolete, low-quality, or context-dependent artifacts, and the process of filtering them requires resource-intensive human oversight. The report suggests that firms will not simply buy AI, because firms will curate themselves for AI, and that curation is labour. The report treats that labour as a major barrier, not as a minor implementation detail.
AI use cases look simple at the surface, but the second category is the strange one
The report groups law-practice use into two buckets. The report describes a first bucket built around automation of repetitive tasks (summaries, prioritisation, first drafts), and the report describes a second bucket that is more conceptually unsettling, because the tool becomes a “creative thought partner” for research, analysis, and innovation. The report argues that the best integration is tandem use, because automation and thought-partner functions reinforce each other when implemented strategically, while the human remains responsible for oversight and final judgement.
The report also signals a risk that many lawyers recognise but rarely name precisely. The report warns that generative output can drift toward the “middle,” because it may lift weak work product while diluting high-quality, distinctive advocacy, and the report insists that outputs should not be treated as final work product or as a replacement for critical thinking. The report sounds almost impatient here, because the danger is not only hallucination, but homogenisation. The report hints that the profession could gain efficiency and lose edge at the same time.
Courts and dispute resolution sit at the fault line between efficiency and legitimacy
The report treats the judiciary as both a beneficiary and a stress point. The report lists possible judicial uses for AI and GenAI (research with caution, drafting routine administrative orders, summarising filings, timelines, editing, scheduling, translation, accessibility services), while the report insists on human verification as a baseline. The report also states that “as of February 2025, no known GenAI tools have fully resolved the hallucination problem,” and the report frames verification as essential for all judicial use cases, not as an optional best practice.
The report also takes deepfakes seriously, not as a pop-culture worry but as an evidentiary and institutional one. The report links AI-facilitated disinformation to threats against the rule of law, and it lists a specific litigation-facing consequence, because judges must handle both deepfakes offered as evidence and strategic claims that authentic evidence is fake. The report treats authenticity and reliability as the core procedural pressure points, not as peripheral issues for “tech courts.”
The report extends the analysis into arbitration and mediation, where the attraction is obvious and the risks remain stubborn. The report describes productivity benefits (summaries, citation linking, pattern detection, comparisons between pleadings, real-time transcription drafts, translations, hyperlinked timelines), while the report highlights familiar risk categories (security, confidentiality, bias, lack of transparency). The report also notes a domain-specific constraint in arbitration, because confidentiality can reduce training data availability, even as specialised databases expand. The report repeatedly pushes the same discipline: independent verification and caution about delegating decision-making to AI agents.
Ethics is no longer one opinion, it is an emerging ecosystem
The report places ABA Formal Opinion 512 (July 2024) at the centre of the ethics story for generative AI. The report summarises the opinion’s message as a requirement that lawyers “fully consider” ethical obligations, including competence, confidentiality, communication, and reasonable fees aligned with time spent using GAI. The report then shows the speed of diffusion, because many state bar associations and judiciaries have issued their own opinions and guidance across 2023–2025. The report reads like a map of professional governance trying to keep pace with software updates, and that mismatch creates a strange kind of regulatory urgency.
The report also anticipates further ethical frameworks in the next two years, potentially including informed consent requirements and transparency duties when AI influences strategy and decision-making. The report connects that movement to confidentiality and data privacy concerns, which are already forcing policy protocols inside institutions. The report treats ethics less as a set of abstract duties and more as workflow engineering with professional consequences.
Access to justice is where AI can help, and where AI can also quietly worsen inequality
The report is optimistic about AI’s potential to expand access to justice, but the report also refuses to pretend that “availability” equals “access.” The report argues that the best way to educate the access-to-justice community is to publicise real use cases from trusted actors, and the report points to work that catalogues and evaluates more than 100 legal-aid use cases. The report also presses for quality standards and continuous assessment, because low-quality tools aimed at individuals without lawyers can cause harm quickly and at scale.
The report then lands on the uncomfortable economic point. The report warns that high subscription costs for reliable legal AI tools can widen the justice gap, because powerful tools could concentrate among clients of means and their lawyers. The report notes some vendor initiatives that offer free or discounted access for legal services providers, while the report insists that financial accessibility must become a standing topic in conversations with developers. The report frames affordability as a governance issue, not as charity.
Legal education is already reorganising around AI literacy
The report describes legal education as a fast-moving laboratory. The report presents survey results showing that 55% of 29 respondent law schools offered AI classes, and 83% offered experiential opportunities like clinics for hands-on learning. The report then gives concrete examples, including a required first-year AI certification at Case Western, AI-driven teaching simulations at Suffolk, an AI Law Lab at Vanderbilt aimed at access-to-justice tools, and a negotiation simulator at Stanford. The report’s subtext is blunt: law schools are training students for a profession whose toolchain will change mid-degree, and faculty are trying to teach adaptability as a professional skill.
Governance, risk, and liability are becoming the new legal industry lingua franca
The report devotes sustained attention to AI governance and risk management. The report summarises a U.S. federal governance track, including OMB Memorandum M-25-21, which requires minimum risk management practices for “high-impact AI,” including pre-deployment testing, AI impact assessments, ongoing monitoring, training, and additional human oversight and accountability. The report also links governance to privacy, civil rights, civil liberties, and discrimination risk mitigation, which is exactly where lawyers are likely to be dragged into technical systems whether they want to be or not.
The report also frames risk categories in a way that is directly usable by counsel. The report lists design and development risks (cybersecurity, privacy, bias, safety, reliability), and the report lists use risks (IP, unfair practices, fraud, disinformation). The report points readers toward the NIST AI Risk Management Framework and the generative AI profile, and the report treats these as foundational governance instruments rather than optional reading for compliance enthusiasts.
The report then adds a telling signal from the insurance market. The report notes the emergence of AI-specific exclusions and restrictions, driven by concerns about “silent AI” coverage, and the report describes a growing market of affirmative AI-specific products covering risks like data poisoning, IP infringement, and regulatory violations. The report effectively says that insurers are pricing the risk even when lawyers are still debating the vocabulary, which should make the profession slightly uneasy.
The report ends by widening the time horizon, and that is the unnerving part
The report closes with a long-term warning that feels deliberately unfashionable in a profession addicted to quarterly deliverables. The report suggests that the most important challenge is not only today’s safe and compliant deployment, but the capacity to anticipate and prepare for a future in which AI capabilities could rapidly jump, leaving institutions “grossly unprepared.” The report assigns lawyers a central role in preserving the rule of law through compliance, risk management, and governance, which is both flattering and burdensome, because it implies that legal institutions may become the last line of defence when technical acceleration outpaces political control.