Zoom announced deepfake risk detection for live video and audio

Zoom used its annual Zoomtopia conference to roll out a new set of AI management features aimed at IT and security teams. The pitch is simple: if AI agents are going to sit inside workflows, companies want a clearer view of what those agents can touch, what they produce, and how much control humans still have over the whole thing.

The biggest change is Centralized AI Management, which gives administrators visibility into AI agents across an organization, along with account-level policies, audit controls, and more control over user-created agents. Zoom says that will arrive in December 2026. The company is also adding custom AI guardrails, so organizations can set rules around what can show up in AI-generated summaries or responses. That can include requiring human review before certain client-facing content goes out, or blocking specific channels from AI processing. That one is expected in Q1 2027.

Zoom is also tying its AI tools more closely to existing compliance systems. Its Sensitivity Label integration with Microsoft Purview syncs labels into Zoom Meetings and Whiteboard, with support for more Zoom Workplace products promised soon. For companies that already classify data in Microsoft 365, that should make it easier to keep those labels consistent as content moves into Zoom. And for regulated industries, Compliance Manager for My Notes and Zoom Canvas is designed to archive AI-generated summaries, notes, and documents; that’s slated for January 2027.

The more eye-catching addition is deepfake risk detection. Zoom says it will analyze live video and audio signals in real time to flag participant video or audio that may have been altered outside Zoom, and it will attach a confidence indicator to the warning. The company is aiming this at hiring, customer interactions, and meetings, where a fake face or voice can create a very ordinary-looking mess very quickly.