We’ve heard the objection twice in recent weeks, both from engineering leaders at regulated financial institutions. “We already have controls,” they say. GitHub Copilot has built-in governance. Our infrastructure team has approval workflows. We don’t need an external layer.

It stops the conversation cold. And it shouldn’t.

Here’s what we’ve learned: the objection isn’t actually about having controls. It’s about what those controls do, and more importantly, what they don’t.

When Matthew Harsh, VP of Software Engineering at Fifth Third Bank, said his Copilot setup had governance covered, we didn’t argue. Instead, we asked: which risks does your current setup actually control for? His answer was specific: seat-level provisioning, usage auditing, maybe IP filtering. That’s real. That’s valuable. Then we asked what happens with prompt injection, data exfiltration through model outputs, model selection decisions across teams, or regulatory reporting when something goes wrong. Silence.

This isn’t theoretical. In regulated environments, controls aren’t just nice to have—they’re required to stay compliant and to demonstrate that compliance under audit. GitHub’s controls are purpose-built for GitHub’s product. They’re not designed for the governance complexity that financial services, healthcare, or insurance face when AI becomes enterprise infrastructure. They cover one layer. They don’t cover all of them.

The second instance followed the same pattern. Another engineering leader at a financial services firm, also confident their existing setup was sufficient. Same question, same discovery: they had controls, but their controls had blind spots.

Two data points are an early signal, not a pattern. We’re not claiming this is universal. But it’s worth paying attention to.

Here’s the frame we’re testing when we hear “we already have controls”: we’re not competing with their current setup. We’re expanding what’s controllable. We ask the prospect to map their risks. Not risks in general. Their specific risks, in their environment, with their regulatory requirements. Then we map what their current controls address. Then we identify the gap.

For Matthew, that gap turned into a meeting Monday at 1:00 PM Eastern. Not because we convinced him his current setup was broken—it wasn’t—but because we helped him see where it wasn’t built to go.

The lesson for us: when a buyer says they already have controls, don’t defend. Lean in. Ask them to show you the map. Ask them to walk through their governance model. Ask them what happens if an engineer accidentally ships a prompt that exposes customer data, or if three teams are using three different models with three different rules. Ask them how they report compliance to audit.

Most will realize their controls are incomplete. Some won’t. Those aren’t the right conversations to have. But the ones who do realize the gap? They’re ready to move fast, because they understand the problem isn’t theoretical anymore. It’s their compliance risk.

That’s the conversation worth having.