Zoom updates terms to train AI on customer data

Nobody reads software terms of service, which is exactly why Zoom thought it could slip its AI ambitions past everyone in the fine print.

According to a report from CNBC on August 7, Zoom quietly updated its terms of service back on July 27 to grant itself broad rights to train its artificial intelligence and machine-learning models on customer data. The move immediately triggered pushback across tech circles.

Here is how Zoom structured the update, and what you are actually giving away.

The opt-out that does not exist

The update carves data into two separate buckets. The first is what Zoom calls "service-generated data."

This bucket covers telemetry, product usage diagnostics, and operational logs. In the terms, users agree that Zoom can access, collect, modify, distribute, and store this data for machine learning and AI model training or tuning.

The catch? There is no opt-out. If you use Zoom, your telemetry feeds its algorithms by default. Tech companies routinely harvest diagnostic logs, but explicitly locking down AI training rights without user recourse shows where Zoom is steering its roadmap.

What about your actual meetings?

The immediate panic from users was whether Zoom's bots were quietly transcribing board meetings and private calls to build its own ChatGPT.

Zoom published a follow-up blog post to clarify the boundaries, stating that audio, video, or chat content is never used to train models "without customer consent."

That last phrase does heavy lifting. In June, Zoom introduced free generative AI trial features, including an automated meeting summary tool and a chat message composer. Turning them on is optional.

However, when administrators or users actually enable these generative tools, Zoom prompts them to sign a consent form that explicitly allows the company to use that exact customer content to train its models. If you want the automated notes, you have to hand over the raw material.

Why it matters

AI companies are running into a massive data wall. With OpenAI, Google, and image generators facing a wave of copyright lawsuits over scraped internet data, enterprise SaaS platforms are sitting on the real holy grail: private, contextual workplace conversations.

In our view, Zoom is playing a high-stakes game with customer trust. Bundling model training permissions into beta productivity features is a clever growth hack for data collection, but it creates a massive compliance headache for enterprise IT departments.

If you discuss confidential roadmaps, financials, or sensitive personal details over video calls, free meeting summaries suddenly carry a steep price tag. For teams handling sensitive information, the smartest move right now is checking your admin dashboard and keeping those generative AI toggles firmly off until Zoom provides clearer data isolation guarantees.