The in-house lawyer who approves an AI-assisted output answers for the decision, whatever tool produced the draft. Courts, professional regulators and boards hold people accountable, not systems. A decision holds when that person had the authority to make it and can show the sources and steps that led to it.
AI now produces first drafts in seconds, conducts contract reviews in the background and compresses an afternoon of research into minutes. Inside the company, it also answers business questions that once came straight to legal. For a GC, this raises a simple but pressing question: once AI has produced the work, who answers for the decision? The system carries neither the authority to make the call nor the legitimacy needed for the board, a regulator or the business to accept it. Both authority and legitimacy remain with the in-house legal team, which is why lawyer supervision of AI is part of the role rather than an extra control.
Key takeaways
- The in-house lawyer who signs off is accountable for the decision.Supervision scales with autonomy: the more a tool does on its own, the more checks legal needs before anything leaves the team.
- Supervision also covers the legal answers business teams take from AI directly.
- A decision holds when legal can show who decided, which sources they used and how to reconstruct the path.
The in-house lawyer who signs off ownsthe decision
The first part of the answer is authority. Whoever signs off is accountable for the decision, regardless of which tool produced the draft.Regulatory guidance is moving in the same direction. California’s 2026 practical guidance on agentic AI states that AI should not make substantive legal determinations, give advice or produce filings without meaningful lawyer supervision, and that controls should strengthen as autonomy increases. The guidance is addressed to lawyers, and in-house counsel carry the same professional duties as their colleagues in private practice. Related amendments to the professional rules are proposed and still open for public comment, its logic travels well beyond one jurisdiction.
Supervision is not a fixed step at the end of a legal workflow; it scales with what each tool is allowed to do. An agent that drafts, routes and sends a regulatory response therefore needs far more checks than one that only suggests wording, and before approval someone in legal has to confirm which steps it took and which sources it relied on. The draft may be accurate, but approval turns it into the company’s position.
Courts apply the same logic. In 2026, a court sanctioned a lawyer over a filing that contained 16 invented quotations. The tool that produced them faced no consequence, and the lawyer who signed did. The setting was litigation, but the principle holds when an in-house lawyer approves a contract, a disclosure or a board paper. The signature, the advice and the certification are human acts, and the liability follows them, which is why accountability for AI-assisted work sits with the legal function.
A decision holds when legal can show its path
The second part of the answer is legitimacy. A decision is accepted inside the company when legal can explain it and show how it was reached.
This matters well beyond the legal team. Business colleagues now put legal questions to AI tools directly, and a persuasive answer is not the same as a legitimate one. Studies have shown that people relied more on AI-generated legal advice than on a lawyer’s when they did not know the source, and could barely tell the two apart. As a result, in-house supervision also means deciding which legal questions the business can settle with AI and which need legal sign-off.
Furthermore, when the stakes are real, people look for the person behind the decision. Research on courts found that decisions by judges using their own expertise were seen as more legitimate than those made with AI or by AI alone. The same expectation applies in the boardroom. If a director asks why the company accepted an indemnity cap in a supplier agreement, someone in legal has to say “I made this call, and here is why”, and that answer holds only if the path behind it can be shown.
A defensible decision therefore rests on three elements: who decided, which sources they relied on and whether that path can be reconstructed later. For a contract decision, those sources are the clause itself, the negotiation record that explains the trade-off and the approval that authorized it. If legal can follow the answer back to each of them, the decision can be explained and defended. An answer with no visible origin, by contrast, asks the board to take it on trust.
Decision traceability: The ability to show who made a legal decision, which sources they relied on and how the path from source to decision can be reconstructed later.
Conclusion
So who answers for the decision once AI produces the work? The short answer is the in-house lawyer who signs it off. That decision holds when legal has the authority to make it and can show the path that led to it. For a GC, supervision therefore covers both the work legal produces with AI and the legal answers the business takes from it.
There is one more condition. The person signing also has to be able to judge what the system produced, and that ability is built over years of doing the work. The second part of this series looks at where that judgement comes from once AI takes over much of the work.
Frequently asked questions
Who is responsible when AI-generated legal work is wrong in an in-house team?
The in-house lawyer who approves the work, relies on it or gives advice based on it remains responsible. Professional guidance in several jurisdictions states that using AI does not reduce or transfer a lawyer’s duties, and in-house counsel carry the same duties as lawyers in private practice. The tool that produced the draft carries no accountability.
How should in-house counsel supervise AI-produced legal work?
Supervision should match what the system is allowed to do. Before approving an AI-assisted output, counsel should confirm which steps the tool took, which sources it relied on and where it filled gaps on its own. The more autonomy the tool has, the more checks the workflow needs before anything is sent, signed or filed.
Should business teams use AI for legal questions without legal review?
That depends on the risk attached to the question. Low-risk, well-documented questions can often be answered with AI inside limits set by legal, while anything that commits the company, such as contract terms, disclosures or regulatory positions, needs legal sign-off. The legal team decides where that line sits and reviews it as tools change.
Does agentic AI change in-house oversight responsibilities?
Yes. Agentic AI can take several steps in a workflow, such as drafting, routing and sending, without a human prompt at each stage. Recent professional guidance on agentic AI states that controls should strengthen as autonomy increases and that legal determinations, advice and filings require meaningful lawyer supervision.
Can legal accountability be delegated to an AI system?
No. An AI system can produce analysis or a draft, but it cannot hold a professional licence, sign off on a contract or answer to a board, a regulator or a court. Accountability stays with the person who decides to rely on the output, which is why that person needs to be able to trace and explain how the decision was reached.
