In the same week this past spring, two narratives about artificial intelligence and lawyer wellbeing landed in the legal trade press, and they could not have been more different. On one side, Clio’s 2026 Legal Trends Report announced that legal technology can reduce cognitive load in legal professionals by up to twenty-five percent, with forty-six percent of lawyers saying that AI made them more likely to stay at their firm for the next two years. A separate Ironclad survey reported that AI was alleviating one of the most stressful aspects of legal work. On the other side, the ABA Journal published Burnout 2.0: AI’s Silent Impact on Lawyer Mental Health, arguing that AI is quietly accelerating the very psychological pressures already embedded in modern legal practice, and the Thomson Reuters Institute reported on Berkeley research suggesting that AI in legal work is producing not relief but recalibrated pressure.
Both stories cannot be true in the way that either is being told. The contradiction matters, because the answer the profession settles on will shape how a generation of lawyers experiences their work — and, more quietly, whether the conditions known to drive lawyer suicide risk are alleviated or intensified.
This article argues that the productivity narrative, while not wrong about what AI can do to a single task in isolation, is wrong about what AI is doing to legal work as a system. When the time required to perform a task collapses, the profession does not bank the saved hours; it absorbs them as a new floor for client expectations, billing targets, and turnaround times. The result is not less work. It is the same work, performed at higher pace, with the cognitive rest that once accompanied rote tasks replaced by a sustained state of evaluative vigilance. That configuration maps precisely onto the predictors of lawyer mental-health risk that researchers have spent the last decade identifying. AI is not the cure for lawyer burnout. On current evidence, it is on track to be a new and more efficient driver of it.
What the productivity narrative gets right
It is worth beginning with what is true. AI tools, used well, can eliminate forms of work that lawyers have long described as soul-crushing. Document review at scale, first-pass contract comparison, discovery culling, citation checking, the assembly of routine correspondence — these tasks have historically consumed hours of attorney time without providing meaningful intellectual reward. When AI does them faster, or does them at all without a human in the loop, the lawyer is freed from a category of work that contributed nothing to professional development and a great deal to exhaustion.
The Clio research is not making the claim up. In carefully designed conditions, lawyers using legal AI tools made fewer mistakes on a will-review task and completed it more reliably than those without. Roughly half of the lawyers in a recent Rev survey identified administrative work as their leading stressor, and AI can plausibly reduce that load. There is a version of the AI story that is genuinely good news for the practicing bar, and it deserves to be told.
The trouble is that the productivity narrative stops at the task level. It does not ask what happens to the profession when every firm in the market completes the same task in a quarter of the time.
Pace recalibration and the expanding sphere of accountability
The legal market is competitive, and competition has a way of converting efficiency gains into baseline expectations. The ABA Journal’s coverage put it plainly: when a lawyer can compare agreements in minutes rather than hours, the profession does not simply admire the improvement, it recalibrates around it. What once counted as fast becomes expected. What once counted as urgent becomes ordinary. Deadlines tighten. Turnaround expectations shrink. Efficiency quietly becomes pressure.
This is not a unique feature of the legal profession. Researchers at George Mason University’s College of Public Health, studying what they call the brain fry phenomenon, found that across knowledge-work sectors, AI did not simply reduce workloads. In many cases it expanded what the researchers describe as the sphere of accountability — workers suddenly felt responsible for producing more work, monitoring more outputs, and managing more information in the same amount of time. The technology that promised efficiency had reshaped expectations to consume the gains.
For lawyers, the sphere-of-accountability dynamic has a particular shape. The associate who once spent an afternoon reviewing fifty contracts now reviews two hundred in the same window — but is also expected to identify edge cases the model missed, evaluate the quality of the AI’s reasoning, flag hallucinated citations, and stand behind the output professionally. The partner who once reviewed an associate’s memo now reviews the associate’s review of an AI-drafted memo, adding a layer of meta-evaluation rather than removing one of supervision. Clients, too, are using AI. A client forwards a chatbot’s analysis and asks why the lawyer’s answer differs. Another sends a draft demand letter generated overnight and expects rapid feedback. The lawyer’s day, formally, is not longer. The cognitive demands within it have multiplied.
The cognitive cost of constant evaluative judgment
There is a specific psychological mechanism at work here, and it is worth naming because the legal-press framing rarely does. Traditional legal software follows predictable pathways: click, complete, move on. AI systems behave differently. The lawyer interpreting an AI output is in a sustained state of active judgment — reading the model’s reasoning, comparing it against legal authority, evaluating confidence, deciding whether to accept, edit, or override. As one industry analysis put it, the brain remains in active judgment mode the entire time.
Cognitive science offers a clear vocabulary for what this costs. Working memory, the system that holds and manipulates information during complex reasoning, has limited capacity. When that capacity is consistently exceeded — by managing multiple AI outputs simultaneously, by maintaining vigilance over systems that produce high volumes of content requiring human judgment — the prefrontal cortex begins to show the effects. Processing slows. Error rates climb. The capacity for complex reasoning diminishes. The lawyer still shows up, still meets deadlines, still bills hours, while operating well below their actual capability.
What makes this particularly hard to detect is that the early stage of AI-augmented work looks like high performance. Output volume rises. Turnaround compresses. Surface metrics improve. The depletion shows up later, and in domains the metrics do not capture: judgment in close cases, willingness to push back on a client, the quiet stamina required to sit with a difficult problem rather than reach for the model. Peer-reviewed research on technostress — the psychological strain associated with rapid technological change at work — has documented statistically significant associations between AI-driven work intensification and both anxiety and depression. The lawyers who feel most efficient under AI may also be the ones whose mental health is degrading fastest.
Mapping AI’s effects onto the known predictors of lawyer mental-health risk
The reason this should worry the profession, and not only the individual lawyer, is that the conditions AI is producing map directly onto the predictors of lawyer suicide risk identified in the best research available. In 2023, Patrick Krill and Justin Anker published Stressed, Lonely, and Overcommitted: Predictors of Lawyer Suicide Risk in the peer-reviewed journal Healthcare. Using a sample of nearly two thousand randomly selected lawyers, they identified four predictors significantly associated with suicidal ideation: high perceived stress, loneliness as measured by the UCLA Loneliness Scale, high work overcommitment, and being male. Perceived stress was the leading predictor. Lawyers with high stress were twenty-two times more likely than lawyers with low stress to have contemplated suicide.
Consider how AI-augmented practice interacts with each of these four predictors.
Perceived stress is, by definition, the felt sense that demands exceed resources. Pace recalibration raises demand without raising resources. The lawyer who once had eight hours to produce a deliverable now has two, and the deliverable is expected to be just as good or better. The objective time pressure increases; the subjective experience of stress follows.
Work overcommitment is the inability to disengage from work — the pattern of thinking about cases at dinner, returning to email in bed, treating work as a category of identity rather than activity. AI tools, by design, follow the lawyer everywhere. There is no closing time for a chatbot. The technology that enables work to happen faster also enables it to happen anywhere, and removes the natural friction that once forced disengagement. Overcommitment is not just enabled by AI; it is incentivized by it.
Loneliness is the dimension the profession is least equipped to talk about, and the one AI may affect most quietly. Lawyers have historically tested their thinking on colleagues — walking down the hall to ask a partner whether an argument holds, sitting in an associate’s office to work through a problem out loud. These exchanges are not just professional, they are relational. They are the small, repeated moments through which lawyers come to know each other and feel known. When the chatbot becomes the first interlocutor, those moments do not happen. The work proceeds; the relationship does not. Over months and years, in a profession where, as the Krill and Anker data show, lonely lawyers are nearly three times more likely to contemplate suicide than their less lonely peers, that erosion matters.
The fourth predictor — being male — is not directly modifiable by AI, but it is worth noting that the cohorts most exposed to AI-augmented work are also the cohorts where loneliness and overcommitment are already elevated. The vulnerability is concentrated where the technology adoption is fastest.
The 2025 ABA-Krill Strategies national research project, with results expected in peer-reviewed publication this year, has explicitly included AI’s impact on lawyer mental health in its design. The profession will soon have national prevalence data on what is currently being inferred from adjacent research. The findings are unlikely to be reassuring.
What a responsible AI practice would look like
None of this is an argument against AI in legal work. The technology is not going away, and a profession that refused it would lose ground it could not afford to lose. The argument is that AI must be implemented with the same care the profession applies to other consequential decisions, and that care must include the psychological health of the lawyers using it.
The Thomson Reuters coverage of recent Berkeley research described what those researchers called an AI practice: a set of intentional norms and routines that structure how AI is used within an organization, including when to stop and how work should and should not expand. The phrase is useful because it borrows the architecture lawyers already understand. A law firm has a practice. The practice has standards, supervision structures, ethical guardrails, and cultural expectations. AI use, at scale, deserves the same treatment.
In practice, that means firms should be asking specific questions. What is the policy on holding the line when AI compresses task time — does the firm reinvest the saved hours in better work, in client value, in supervision, or simply in more matters per attorney? What expectations are being communicated to clients about turnaround, and are they sustainable? What is the firm’s posture on AI use after hours, on weekends, during vacation? Are associates being supervised in their use of AI, or are they being left alone with tools whose outputs they are not yet equipped to evaluate? Is the firm tracking not just productivity metrics but the indicators that actually predict mental-health risk — sleep, weekend work hours, isolation, alcohol use, the felt sense among lawyers that the profession is harming them?
These are not soft questions. They are the questions that distinguish a firm whose AI strategy serves its lawyers from one whose lawyers serve its AI strategy. The early evidence suggests the second model is the default, and that intentional countermeasures are required to avoid it.
The question is not whether, but on whose terms
The legal profession is not the first knowledge-work industry to confront this dynamic, and it will not be the last. What is particular to lawyers is that the profession already carries elevated rates of anxiety, depression, problematic drinking, and suicidality compared to the general working population. The margins for absorbing additional psychological strain are thinner here than almost anywhere else. A change that might be merely uncomfortable in another industry can be genuinely dangerous in this one.
The right question is not whether to use AI. It is on whose terms — the firm’s, the client’s, the market’s, or the lawyer’s. So far, the productivity narrative has been winning, in part because it is easier to measure billable efficiency than psychological strain, and in part because the strain takes longer to surface than the efficiency. By the time the ABA-Krill data lands, a great deal of the recalibration will already have happened.
Attorneys who are noticing these patterns in their own work — the harder time disengaging, the sense of constant evaluative pressure, the missing relationships with colleagues, the quiet fatigue that survives a weekend — would do well to take those signals seriously rather than dismiss them as the cost of adopting a new tool. They are signals worth bringing to a therapist who understands the legal profession. AttorneyTherapists.com maintains a directory of licensed clinicians who specialize in working with attorneys, for readers who want to find one.
