Who owns AI at a law firm?

How we know this

Sidebar puts one question a week to legal management professionals at firms of 10 to 200 attorneys. Members are verified by title, employer, and firm size before they are admitted, and every reply is private. This page draws on every Sidebar cycle that has touched this question, and it is updated as new replies come in. We publish patterns across the group, never individual firms, and only once at least five members have replied to that question. Full methodology at gosidebar.ai/methodology.

Committees exist; ownership doesn't

Most mid-size firms have built something that looks like AI governance: a committee, a charter, a recurring slot on the agenda. Formal structure is not the same as accountability. A committee can meet for months and still leave the question of who decides unanswered, because meeting on schedule and deciding something are two different disciplines, and only one of them shows up automatically on a calendar invite. When the group stalls, the work does not wait. It defaults to whoever cares most, which is rarely the person the org chart would name. The pattern repeats often enough to look less like an accident and more like a design flaw: a committee is easy to stand up and hard to hold responsible, which makes it the path of least resistance for a firm that wants to look like it is addressing AI without anyone having to own the outcome.

COOs are absorbing what nobody else will

Legal management professionals keep ending up as the de facto owners of AI, not because anyone chose them, but because nobody else stepped up. IT keeps scope narrow, treating AI as one more application to license and patch rather than a decision that touches how the firm practices law. Partners are too busy to drive it week to week, and the people with the standing to decide are rarely the ones with the time to sit through vendor demos. So the COO steps in: evaluating vendors, setting the guardrails, fielding the questions nobody else will answer, and absorbing work that a formally assigned owner would otherwise carry. Ownership by accident, and more common than anyone planned for. It holds up reasonably well given who is in the room week to week, and it is fragile in a specific way: the role was never granted, so it carries none of the authority that would let its holder tell a partner no.

Partnership sign-off is the universal bottleneck

Regardless of who evaluates a tool or drives the day-to-day work, 58% of firms route the final call through full partnership approval. That structure manages to slow adoption and diffuse accountability at the same time. A vote spread across the full partnership means no single person owns the outcome, and it adds a step that a fast-moving technology decision can rarely afford to wait on. The real bottleneck at these firms is not that too few people weigh in. It is that everyone does, on a decision only a handful of them are equipped to evaluate. Full partnership sign-off reads as caution, but caution and diffusion are not the same thing, and a decision nobody owns is not more careful than one a single accountable person makes. It is simply slower to arrive at, and harder to trace back to a source when it turns out wrong.

Informal champions carry disproportionate weight

In smaller firms especially, a single motivated partner or executive turns out to be the real engine behind AI progress, whatever the org chart says about committees or IT. That person pushes because they care, not because a role description asked them to, and the firm's progress tracks their attention almost exactly. When that person is unavailable, pulled onto a matter, on leave, or simply focused elsewhere, momentum stalls entirely, because nothing underneath them was built to keep moving without them. The dependency stays invisible right up until it gets tested. A firm can spend a year believing it has a functioning AI program, when what it actually has is one person's sustained interest, and the two look identical until that person stops paying attention. The uncomfortable part is that this arrangement often looks like success from the outside. Tools get evaluated, licenses get signed, questions get answered, and none of it required a policy or a title. It only required someone who cared enough to keep showing up, which is exactly the resource a firm has the least ability to plan around.

Evaluation rigor is inconsistent

About 26% of firms run structured, side-by-side testing: the same data through multiple tools, comparing outputs before committing budget. Most do not. Most decide on gut instinct, a strong vendor demo, or early feedback from whoever tried the tool first, which is a reasonable way to move fast and a poor way to defend a renewal decision a year later. The gap between the two groups is not sophistication. It is discipline. Structured testing costs time up front that a firm under pressure to show progress rarely feels it has, and the payoff, a defensible answer when someone asks why this tool and not another, does not show up until the decision is already being questioned. Firms that build the habit early have an answer ready when the question comes. Firms that do not are relying on the same instinct that produced the last tool nobody opens anymore.

The approved list is not the ownership boundary anymore

The ownership gap above is showing up in a new place: tool sprawl. Firms report several AI tools running at once with no single person coordinating between them, and the tool formally on the approved list is often not the one people reach for day to day. That is a harder ownership problem than a stalled committee, because it means whoever holds nominal ownership of the firm's AI tool has no visibility into what the firm's people are really using, which is scattered across whatever assistants individuals already trust. Firms naming this are starting to ask for a single consolidated environment and a formal process to close the gap between what is sanctioned and what is running underneath it, which is itself a tacit admission that current ownership, wherever it formally sits, has not been reaching far enough to see the problem it is supposed to own.

The public record shows the same gap, at a larger scale

The 8am 2026 Legal Industry Report, drawn from more than 1,300 legal professionals surveyed in the fall of 2025, found that 43% of firms have no formal AI policy at all and no plans to write one, the most basic form of documented governance a firm can put in place. Sidebar members skew ahead of that baseline; most describe some kind of committee or approval process already in place. But the wider number is worth sitting with, because it means a large share of the market has not even reached the point where accountability could break down. They never assigned it in the first place. Our members' problem, a committee that exists but cannot decide, is in some ways the more advanced failure mode. It is what accountability avoidance looks like once a firm has cleared the paperwork stage and still has not answered the actual question of who is on the hook.

Source: 8am, 8am 2026 Legal Industry Report

What to do with this

Name one accountable owner of AI decisions this quarter, not a committee, an actual name attached to the hard calls. A committee is a socially acceptable way to make AI someone else's problem, and firms that lean on one are choosing the appearance of governance over the substance of it. If a motivated person is already carrying that weight informally, make it formal: give them the standing to tell a partner no, instead of only the workload of fielding questions nobody else will answer. Then build the part almost nobody has: a second person who knows enough of the landscape to keep the program moving if the first one leaves, goes on leave, or simply burns out. Write down the handful of decisions that person is authorized to make alone, and the handful that still need a wider conversation, so the boundary of the role does not have to be renegotiated every time something hard comes up. Ownership by accident is not a scalable strategy, and right now most firms have nothing underneath the one person currently holding it together. The firms still running a functioning AI program two years from now will not be the ones with the most committees. They will be the ones that turned an accidental owner into a designated one, and gave that person someone to hand it to.

Frequently asked questions

Who is responsible for AI decisions at a mid-size law firm?

Formally, it is often a committee, a technology partner, or IT, but real accountability tends to land somewhere else. Across mid-size firms, 37% report no one is willing to own the hard calls, and COOs are increasingly stepping into the role by default rather than appointment.

Should a law firm's AI decisions go through a committee?

A committee can work, but only if it can say no, including to a senior partner. About 58% of firms route final AI decisions through full partnership approval, a structure that slows decisions and spreads accountability so widely that no one person owns the outcome.

What happens when a law firm has no clear AI owner?

The work defaults to whoever cares most, usually the COO or another operations leader, rather than to someone formally assigned. That arrangement holds until the informal owner is unavailable, at which point momentum tends to stall because nothing was built to run without them.

How many law firms test AI tools before buying them?

About 26% run structured, side-by-side comparisons using the same data across multiple tools before committing. Most decide on a strong vendor demo or early feedback from whoever tried the tool first, which moves faster but leaves little to point to when a renewal decision gets questioned.