Three AI Governance Signals Every Legal IT Leader Should Take From ILTACON 2026

Every year, an industry conference tells you less about the event itself and more about where an entire market’s anxiety is pointed. This year, ILTACON, one of the legal industry’s largest technology gatherings, made that anxiety impossible to miss. Conversation after conversation circled back to the same theme: enterprises are deploying AI agents faster than they can govern them, and the gap between the two is starting to show.

Legal is a useful bellwether here. Law firms and corporate legal departments operate under some of the strictest confidentiality, ethics, and audit obligations of any industry, which means their AI adoption problems tend to surface earlier and more visibly than they do elsewhere. What legal IT leaders are grappling with today is a preview of what most enterprises will be grappling with within the next 12 to 18 months.

Here are three signals from the show floor that matter well beyond the legal sector.

1. AI Agents Are Outrunning AI Governance

The most consistent theme at ILTACON wasn’t a specific product or vendor. It was a shared realization: organizations no longer have a clear, current picture of what AI is actually doing inside their environment.

This isn’t a hypothetical risk. Employees connect personal AI tools to corporate systems. Vendors quietly ship AI features into software that’s already been approved and deployed. Developers stand up custom agents that never go through a formal review process. Each of these is a reasonable, well-intentioned decision made at the individual level. Collectively, they add up to an AI footprint that IT and security teams can’t fully see, let alone govern.

The old governance model, built around periodic audits and static documentation, assumes you know what you’re governing. That assumption breaks down when new agents, integrations, and AI-enabled features can appear inside your environment without a formal deployment event. Governance has to become continuous and automatic, not a quarterly exercise.

For enterprise leaders, the practical question isn’t “do we have an AI policy.” Most organizations do. The real question is whether that policy can be enforced against AI activity you haven’t specifically catalogued yet. Discovering every agent, model, and tool running across your environment, including the ones nobody formally approved, is quickly becoming a prerequisite for governance rather than a nice-to-have.

2. MCP Adoption Is Outpacing Its Security Model in Legal

The Model Context Protocol, or MCP, came up constantly in conversations at ILTACON, and for good reason. Legal teams are connecting AI agents to document management systems, e-discovery platforms, and matter systems faster than most IT departments can formally review those connections. MCP has become the default way agents reach those tools, and its flexibility is exactly what makes it both useful and a governance headache.

Two concerns surfaced repeatedly on the show floor. The first is security: every MCP connection is effectively a new access point into systems holding privileged, client-confidential data, and a growing number of those connections are being stood up without centralized oversight. The second is cost: firms experimenting with MCP are discovering that ungoverned agent-to-tool connections can quietly inflate token consumption in ways that are difficult to trace back to a specific matter or practice group, complicating both budgeting and client billing.

Both problems share the same underlying fix: centralizing how agents connect to tools, rather than letting each practice group or vendor wire up its own point-to-point integration. A governed gateway approach lets IT and security teams enforce policy consistently across every connection and preserves the audit trail firms need for both client obligations and regulatory readiness.

Airia recently expanded its MCP Gateway with new routing and access control capabilities, including semantic tool search that keeps large tool catalogs from overloading a model’s context window, and role-based access controls that route each user to only the tools their role requires. As firms connect more systems and more users to MCP, that kind of granular control becomes essential rather than optional.

If your firm is experimenting with MCP, or already relying on it without a formal governance program, this is the moment to get ahead of it. The pattern from ILTACON was consistent: firms asking about MCP governance now are in a much stronger position than the ones who wait until an incident, or an unexplained token bill, forces the conversation.

3. ISO 42001 Is Becoming the Practical Standard, Not the EU AI Act

If you’re tracking AI regulation, most of the public conversation over the past two years has centered on the EU AI Act. That wasn’t what dominated conversations at ILTACON. ISO 42001, the international standard for AI management systems, came up far more often, and with far more urgency.

The reason is practical. The EU AI Act applies to a specific set of organizations under a specific set of conditions, and enforcement timelines have been a moving target. ISO 42001, by contrast, is something organizations can pursue certification against right now, and it’s increasingly being requested by customers, partners, and procurement teams as proof that an organization has a real AI management system in place, not just a policy document.

Multiple attendees described their organizations as actively preparing for or beginning ISO 42001 audits. That preparation work, building an inventory of AI systems, documenting risk classifications, establishing controls, and generating audit-ready evidence, is exactly where most organizations discover how far their governance maturity lags behind their AI adoption.

The organizations that treat this as a documentation exercise tend to struggle. The ones that build governance and control directly into how their AI actually runs, so that compliance reporting is generated from real system activity rather than assembled after the fact, are the ones walking into audits with confidence instead of dread.

What This Means for Enterprise Leaders

Taken together, these three signals point to the same underlying shift. AI governance is moving from a policy conversation to an operational one. It’s no longer enough to have a well-written AI use policy sitting in a compliance folder. Enterprises need real-time visibility into what AI is running, enforcement that happens at the point of action rather than after a review, and documentation that’s generated automatically rather than reconstructed under deadline pressure.

A few practical starting points, regardless of industry:

• Inventory before you enforce. You can’t govern AI activity you can’t see. Start with a real, current picture of every agent, model, and tool in use.
• Treat MCP and agent connectivity as a security surface, not just a productivity feature. Centralized, policy-enforced connections beat ad hoc integrations every time.
• Build toward ISO 42001 readiness now, even without a certification deadline. The documentation and control discipline it requires is good practice on its own merits.
• Assume governance needs to be continuous. Annual audits and static policies can’t keep pace with how quickly AI capability shows up inside an organization.

Legal is simply where these pressures are showing up first. Every enterprise running AI at scale will face the same questions ILTACON attendees were asking this year: what’s actually running, who approved it, what can it access, and could you prove any of that under audit today.

Organizations that start building the answer now will be in a fundamentally different position than the ones scrambling to reconstruct it later. See how Airia can help you take control and govern your entire AI ecosystem today. Connect with a member of our team to get started.