Your GC is already asking how you’d prove that an AI agent didn’t pull from a restricted engagement when it answered a partner’s question. And most firms don’t have a clean answer but because the architecture required to answer it confidently wasn’t part of the original deployment decision.
That gap is more manageable than it seems once you understand where it exists, and closing it unlocks more than just a defensible answer to the GC. According to Intapp’s 2026 Tech Perceptions Survey, 76% of professionals are using unauthorized AI at work – even at firms that have already deployed enterprise AI. Official adoption alone doesn’t stop the behavior that creates exposure. Generic enterprise AI governance was never designed to uphold the professional obligations that regulators and clients expect from firms. The firms that built the right governance architecture first are the ones scaling AI across the engagement lifecycle rather than running pilots indefinitely.
“Trust is really what determines whether or not AI can scale,” said Brant Hollenkamp, Professional Services Industry Leader at Microsoft. “If leaders can’t see what AI is doing, they can’t govern it, and they can’t measure the outcomes — adoption will stall.”
Trust is really what determines whether or not AI can scale. If leaders can’t see what AI is doing, they can’t govern it, and they can’t measure the outcomes — adoption will stall.
Brant Hollenkamp, Professional Services Industry Leader at Microsoft
The check your agents aren’t running
Microsoft Purview governs file access by applying sensitivity labels to documents with information barriers at the user and group level. For most enterprise environments that’s sufficient, but a professional services firm running AI agents across its engagement data is not a standard enterprise environment.
When a partner queries an agent about a target company’s financial position, the agent checks whether that partner has access to the files it wants to surface. It does not check whether that partner has a live restriction tied to that company or a personal shareholding that should have excluded them from the engagement entirely. Those checks live in your compliance systems – and that gap doesn’t close by switching models. Any agent querying your data will return whatever the underlying access controls permit regardless of which model is running.
That distinction matters the moment a regulator or independence reviewer asks a firm to prove a specific partner never had access to restricted engagement data. An access log shows what files were labeled. It cannot reconstruct whether a conflict existed before the query ran – and that is the question they are asking.
“Purview was always designed to prevent you gaining access to information,” said Richard Bowes, Senior Compliance Growth Director at Intapp. “It’s not really designed to prove the negative.”
How Intapp and Microsoft govern every agent
The firms moving fastest with AI are doing so because they invested in governance. A Grant Thornton survey found that 78% of executives say they are not prepared for a governance audit and 46% specifically cite data governance failures as the primary reason. The firms that have built the governance foundation have removed the hesitation that stalls everyone else — a firm that can prove engagement-level controls are functioning doesn’t have to re-litigate the risk question each time it deploys a new use case.
When engagement data is governed at the engagement level, the same foundation that protects independence enables the firm to teach its AI what the firm knows. Key learnings from closed engagements can be anonymized and structured so that the AI surfaces relevant precedents when a team is scoping a new mandate without crossing the walls protecting active client data. A partner walking into a first meeting with a manufacturing sector target can ask their firm’s AI what the last four engagements in that sector found and receive a structured briefing built from the firm’s own work. That’s not incremental efficiency. That’s the kind of institutional intelligence that has historically walked out the door with every senior manager who moved on.
When you can teach your AI and inform your AI with your firm best practices, now that’s gold.
Maria Volokh, Risk Management and Compliance Leader at Intapp
Intapp Walls makes that foundation possible by attaching governance to the engagement data rather than configuring it separately for each AI tool the firm deploys. The rules it enforces are the same independence and conflicts rules already in place for standard operations, so every agent the firm adds operates within the same boundaries from day one.
It also answers the question every firm raises about adding AI governance: whether it means managing yet another standalone system. Walls extends Purview’s reach into the engagement context it was never designed to capture rather than running as a separate system alongside it. And as the firm deploys more agents, coverage expands with every new tool rather than creating a new gap.
Before you add the next agent
The right time to invest in engagement-aware governance was before broad deployment, but the more useful question now is what must be true before the investment pays off.
Three prerequisites determine readiness:
- Engagement data needs to be structured and tagged at the engagement level rather than the file level
- Tagged engagement data only protects the firm if the logic behind it is machine-readable rather than locked in a system that agents can’t query
- When both of those are in place, the final requirement is an audit trail that captures not just what the model returned but what data it was permitted to access when it ran
Firms that have those three things in place aren’t just protected. Every governed engagement adds to the knowledge base, and every anonymized key learning makes the next pursuit sharper than the last. The firms building that foundation now will be the hardest to catch. The firms still waiting will get there eventually, but they’ll get there later and with less to show for the time spent running AI pilots.
“The firms being really successful with AI are the ones that have developed this concept of an AI factory,” said Brant Hollenkamp, Professional Services Industry Leader. “They have to figure out how to get to fifty, to one hundred, to one hundred and fifty and even more use cases.”
The firms being really successful with AI are the ones that have developed this concept of an AI factory.
Brant Hollenkamp, Professional Services Industry Leader at Microsoft
Watch Governed AI in action: how Intapp and Microsoft enable trusted AI for professional services firms to learn how engagement-aware governance models close the gap.