For the better part of three years, the legal industry's conversation about AI competence has orbited around professional responsibility: what Model Rule 1.1 requires, what state bar opinions recommend, what CLE credits attorneys should accumulate. That framing, while not wrong, has obscured the more immediate threat. The exposure most law firms face from undocumented AI use is not disciplinary. It is financial. Fee arbitration panels, malpractice carriers, and federal courts are beginning to treat AI-assisted work without adequate documentation as grounds for fee reduction or disgorgement. The audit trail, not the training certificate, is what firms need to produce.

The Fee Challenge Playbook Is Changing

Opposing counsel in fee-shifting litigation has always had incentives to scrutinize hours billed. What is new is the tool they are reaching for. AI usage disclosures are now a recognizable litigation tactic in fee challenges, and the documentary hooks to pursue them are proliferating. By the first quarter of 2026, at least a dozen major jurisdictions have issued standing orders requiring disclosure of AI use in submitted work product. These orders were designed to promote accuracy and accountability. Their secondary effect is to create a paper record that fee challengers can subpoena, examine, and argue from.

The trajectory here matters. The 2023 sanctions in Mata v. Avianca were widely reported as a cautionary tale about hallucinated citations. What received less attention were the downstream fee consequences: hours spent producing defective work product became difficult to defend, and the episode demonstrated how quickly AI-related failure converts into a billing dispute. That case was an early and particularly stark example. The pattern it illustrated has since become more routine and more strategically deliberate.

Clients, too, are evolving their positions. Sophisticated in-house legal departments, accustomed to scrutinizing outside counsel invoices, are beginning to ask whether billed hours on research and drafting tasks reflect appropriate use of available AI tools. The question cuts two ways: a firm that billed forty hours on a research task that AI could have compressed may face a reasonableness challenge on the hours; a firm that billed AI-assisted hours without documenting the human review layer may face a challenge on the value. Neither position is comfortable without documentation.

What the Lodestar Inquiry Now Absorbs

The lodestar method, the dominant framework for court-assessed fee petitions, multiplies reasonable hours by a reasonable rate. Courts have long scrutinized what counts as reasonable hours. What is beginning to happen is that the reasonableness inquiry is absorbing a new question: would a competent firm have used AI for this task, and if so, what does that mean for the hours claimed?

There is no settled doctrine here, and that ambiguity is itself the risk. Courts assessing fee petitions have broad discretion. A federal district judge reviewing a fee petition in a civil rights case or a securities matter is not constrained by a clear rule when evaluating AI-assisted research hours. What the judge does have is intuition, and increasingly, information. As AI tools become more widely understood, the assumption that document review or legal research consumes the same attorney-hours it consumed five years ago is becoming harder to sustain without a record showing what actually happened.

Firms that can produce that record are protected. Firms that cannot are exposed to reductions that are, in practice, impossible to appeal effectively because they rest on the court's discretionary assessment of reasonableness. The absence of documentation does not just create vulnerability; it forfeits the argument.

The Insurance Market Is Pricing This Before Bar Discipline Does

Malpractice carriers are rarely early movers on professional liability issues. They follow loss experience, and loss experience follows events. The fact that several major legal malpractice insurers have already begun adding AI governance questions to renewal questionnaires is significant precisely because it suggests the market has seen enough early signals to begin pricing the risk.

The questions appearing on renewal forms are specific: Does the firm have a documented AI governance policy? What vetting procedures apply to AI tools before deployment? Do attorneys receive AI-specific training? Firms without satisfactory answers to these questions are seeing premium adjustments. Firms with documented frameworks are not being rewarded yet, but the differential is likely to widen as the claims docket develops.

This is the insurance market doing what it does: quantifying risk before regulators codify it and before courts establish precedent. Bar discipline for AI-related competence failures remains relatively rare. Malpractice claims arising from AI-assisted errors, however, do not require bar action to generate liability. A carrier that pays a claim on a matter where defective AI output contributed to the error has every incentive to reconsider underwriting terms. Firms that have not closed the documentation gap are underwriting that risk themselves.

The Operational Gap That Most Firms Have Not Closed

The characterization of law firms as reckless adopters of AI does not hold up. Most large and midsize firms have adopted tools through deliberate processes, engaged vendor representations, and conducted at least some attorney training. The failure is elsewhere. It is in the infrastructure that would allow a firm, when a fee challenge arrives eighteen months after the work was done, to reconstruct precisely who reviewed an AI-generated output, when that review occurred, what changes were made, and what independent judgment was applied.

That reconstruction is currently impossible at most firms. Use logs do not exist in retrievable formats. Review records are not systematically maintained. Output audit trails were never created. This is not a cultural problem, a problem of attorneys using AI carelessly or without thought. It is an operational problem: the tools being used are not built to produce defensible records, and the workflows around them were not designed with a fee challenge in mind.

The distinction matters because it points to the solution. Attorneys cannot be trained into producing documentation that their tools structurally cannot generate. The infrastructure has to support the record-keeping. Asking associates to manually document every AI interaction is both unworkable and insufficient. The documentation has to be embedded in the workflow itself.

Purpose-Built Infrastructure Versus Adapted General Tools

This is the functional distinction that the market has not yet fully absorbed: AI used in a legal context is not the same as AI built for a legal context. Consumer large language models and general-purpose productivity tools, even when capable and accurate, are not designed to produce the version history, reviewer attribution, research provenance, and human-in-the-loop documentation that fee disputes require. They generate outputs. They do not generate records.

Platforms purpose-built for legal workflows approach the problem differently. Version control is inherent, not optional. Reviewer attribution is embedded, not reconstructed after the fact. Research provenance is traceable. When a fee challenge asks how a document was produced and by whom it was reviewed, the answer exists in the system rather than in the memory of a third-year associate who may no longer be at the firm.

This infrastructure layer is the practical response to the audit trail problem. It is also the response to the insurance underwriting problem, because a documented governance framework built on purpose-built tools is precisely the evidence that renewal questionnaires are probing for. The two exposures share a solution.

Transparency as a Commercial Position

There is a forward-looking argument here that goes beyond risk mitigation. Forward-thinking firms are beginning to include AI usage summaries in client reports and engagement documentation, not as a defensive measure, but as a transparency signal. Sophisticated in-house legal departments, which have their own reporting obligations and their own exposure to scrutiny over outside counsel spend, are starting to ask for this information. Firms that can provide it are differentiating themselves on a dimension that fee-only competition cannot replicate.

The firms that will occupy the strongest position in this environment are those that treated the documentation question as a governance design problem before it arrived as a litigation problem. The ones that deferred governance until the first fee challenge will spend the intervening period absorbing risk they need not have carried.

The competence standard under Rule 1.1 has not changed. What has changed is the evidentiary context in which competence is assessed. Courts, clients, and carriers are all, through different mechanisms, asking the same question: can you show me what happened? Firms that have built the infrastructure to answer that question are not merely protected. They are positioned. The audit trail has become the argument.