In-House Legal Team Reactions to Outside Counsel AI Claims

In-house legal teams have adopted the technology faster than firms and bring their own operational experience to pitches and invoices. Their central test is not whether a firm has AI tools, but whether it can show governed use on actual matters, validated work, protected client data, and fees that reflect how the work was done.

Why in-house legal teams now evaluate outside counsel AI claims from a position of strength

Most corporate legal departments now use AI, but law firms are behind in rolling it out across the firm and training people to use it. That difference changes the starting point of a conversation about outside counsel capability. The firm is no longer necessarily introducing a technology that the client has yet to understand. Instead, the general counsel may already have overseen its rollout, checked its output, and seen where people still need to review the work. The client therefore has a benchmark independent of the pitch.

FTI Consulting/Relativity reports that legal departments with formalized technology roadmaps have more than doubled year-over-year. A roadmap places AI within planned operations. It also gives legal teams a basis for comparing a firm's ambitions with their own implementation work. That hands-on experience helps them even though they may not know every tool or way to use it in legal work. GCs know enough about deployment to distinguish access to software from a working professional capability.

What drove in-house adoption faster than firms — budget pressure, not enthusiasm

The Thomson Reuters Institute Law Firm Rates Report 2026 says law firm rates are rising at roughly twice the rate of inflation. For in-house departments, that pressure comes out of a legal budget that must also meet business demand. They cannot recover the cost of an inefficient internal process by billing more hours for it. AI gives them a way to handle routine work without turning to outside counsel who cost more and more. Faster adoption follows from that constraint, rather than necessarily from greater enthusiasm for technology.

Law firms face a different short-term calculation because time saved can mean billable time lost. Partners billing at top-of-market rates have little structural incentive to argue that their work has become faster and cheaper. That helps explain why the two sides moved at different speeds. In-house lawyers gained operational experience while firms still had to reconcile efficiency with their revenue model. That divergence now sits beneath both the capability discussion and the pricing discussion.

What in-house teams are actually doing with AI — the baseline GCs bring to every pitch

ACC/Everlaw identifies drafting and legal research as the most common in-house AI applications. FTI Consulting/Relativity adds summarization, contract clause identification, and first-pass document review. These are core legal workflows in which users can compare an AI-assisted result with the work it replaces. They also expose the limits of automation because an initial answer is not necessarily a reliable final product. The experience gives GCs a concrete basis for judging claims about speed, scope, and review.

Salesforce's internal legal-ops team deployed an AI agent to draft, redline, and analyze contracts autonomously. The work previously required billable outside counsel hours, and the deployment reduced total spend by millions of dollars. LegalZoom used GenAI for a different, tightly bounded purpose, deploying an internal AI call simulator for sales agent training. It achieved measurable productivity gains quickly by limiting the project to a single workflow.

Tiffany Lee, general counsel of Liquid Death, now tells outside counsel they must use AI for routine work and the first round of research. Her four-person legal team consists of two in-house attorneys, one contract attorney, and one contract paralegal. She reports that the “gut check” calls she used to make will happen less as the team handles verification internally. That change affects not only how work gets done, but which work reaches a firm at all. A GC with this experience can assess an outside counsel proposal against an internal alternative, including the time needed to check the result.

The transparency gap — why most GCs don't know whether their outside counsel is using AI at all

ACC/Everlaw identifies a “transparency gap” in which nearly six in ten in-house counsel do not know whether their outside counsel is using GenAI on their matters. That uncertainty is distinct from doubt about whether AI can help with legal work. Clients may know the technology well but know little about how a given firm uses it. Without that information, they cannot connect a capability claim to the service they receive. The missing link is disclosure at the matter level.

The gap is structural rather than necessarily evidence of deliberate concealment. Many legal departments have no formal position on outside counsel AI use, so expectations may never have been settled between the parties. Thomson Reuters CEO Steve Hasker has noted that articulated AI strategies are not reflected in teams' day-to-day experience. A firm may consequently struggle to describe its own use, leaving the client to guess whether AI-assisted work is being billed as though the process had not changed.

How the transparency gap becomes a billing dispute

That uncertainty starts to affect fees when a familiar task takes much less time. A contract review that once took a mid-level associate well over an hour may take a fraction of that time with legal AI assistance. Yet an invoice that does not explain the method can look much the same to the client. An experienced GC may see how much time could be saved without being able to tell whether the bill takes that into account. The dispute is then about the relationship between effort and price.

Firms have good reason to question the view that faster generation makes legal work cheaper. AI work on sensitive matters needs to be checked, and using it with care may cost more because of hallucination and privilege risks. An initial draft or research answer does not eliminate the professional work needed to stand behind it. That objection deserves weight when assessing a particular task and its fee. It becomes self-serving, however, when used as a blanket defense against disclosure, because review requirements do not explain why a client cannot know whether AI was used.

Privilege also imposes a real constraint on the choice of deployment. Public consumer-tier LLM products can put privilege at risk when client data is used for training, while most major LLM APIs keep customer data out of training by default. Legal enterprises are moving toward private deployments hosted in isolated cloud environments in response to these concerns. That option is available to firms as well as in-house departments. The relevant distinction is therefore between specific data protections and exposures.

What GCs are specifically scrutinizing in outside counsel AI pitches

GCs first examine whether a firm's account of AI use reaches the client's matter. A general policy does not establish which tasks were AI-assisted or how that assistance relates to an invoice. Experienced buyers look for a connection between what the firm says it can do and what it actually does on an engagement. They also distinguish a lawyer trying a tool alone from a firm using it in a governed process to deliver work.

Stanford RegLab's peer-reviewed study found that leading legal research tools hallucinated at a significant rate. A GC probing how a firm validates output is therefore testing an operational risk. Data handling brings a similarly concrete inquiry into whether deployments are private and tenant-isolated and keep client data out of model training pipelines. Neither issue is answered by saying that a tool was built for lawyers.

GCs also look at pricing because the value of time saved depends partly on how the firm charges for the work. ACC/Everlaw reports that 61% of in-house counsel are likely to push for alternative fee arrangements. Fixed or capped fees become relevant where AI has made categories of work faster and effort more predictable. GCs also distinguish a partner reviewing an in-house-drafted position from a partner developing that position from scratch.

Why firms so often fall short when GCs probe these questions

The 8am Legal report finds that a majority of firms have not deployed AI at the organizational level and provide no training. A large share also have no formal AI policy. Those gaps leave lawyers to act on their own when clients may assume the firm supports their work. They help explain why a persuasive demonstration can be followed by weak answers about validation, governance, or billing. The tool may work, but the organization has not established how to deliver it as a dependable service.

Thomson Reuters CEO Steve Hasker describes the wider implementation problem in similar terms. Expected improvements in quality, speed, and efficiency have not materialized because organizations have not changed around the technology. In a law firm, leaders may back a strategy before it becomes part of the work on individual matters. A demonstration establishes that a tool can perform a task under the conditions shown. It does not establish that the firm has put that capability into governed production.

Client procurement has not yet imposed consistent pressure to close the gap. Only a small share of in-house teams have put formal GenAI provisions into RFPs or outside counsel guidelines, so many claims go unchecked against contract terms. At the same time, only roughly a fifth of legal AI users report high trust in AI output. That limited trust makes the absence of validation workflows more consequential. The failure is organizational and fixable, but a firm cannot stand behind AI-assisted work merely by pointing to its lawyers' access to a tool.

What credible AI use actually looks like from the client side

Credibility begins when a firm can describe to the client which tasks receive AI assistance and how the resulting work is reviewed. Documented validation gives substance to that account without relying on assertions that a tool is “hallucination-free”. Clear accounts of how private, tenant-isolated, no-training-data environments work also show clients how their data is handled.

GCs who defined success criteria before their own deployments and tracked cycle time and error rates will bring that lens to outside counsel. Login statistics do not answer whether legal work has become faster or more reliable. The AI Operations Audit framework asks whether AI is used as a governed professional workflow tool or an untracked general-purpose assistant. Where governed use makes effort more predictable, credible firms open the discussion about fixed or capped fees rather than leaving the efficiency disconnected from price.

The durable relationship is not one in which in-house teams and outside counsel compete to automate the same work. The 2026 CLOC Global Institute discussion points toward in-house teams handling AI-assisted first passes while firms contribute complexity, judgment, and scale. Jasmine Singh of Ironclad framed this as using each party for what it is specialized for. Credible AI use supports that relationship by making the division of work visible, the output defensible, and the economics intelligible to both sides.