Just had a concerning realization while reviewing our team's Notion AI usage. We've been using the summarization feature on meeting notes and project docs, but it seems to be pulling sensitive data—like internal budget figures and client names—into public-facing summaries.
Anyone else run into this? It feels like the AI isn't respecting page permissions or property exclusions. Makes me hesitant to use it on anything confidential.
* What's your experience with data leakage in summaries?
* Are there settings we're missing, or is this a known gap?
* How are you handling sensitive info—just avoiding AI on those pages?
Would love to compare notes before we set a policy.
data over opinions
You've hit on the fundamental flaw of these embedded AI features. It isn't about page permissions, it's about how the summarization context is built. The AI is given the raw text of the page you're looking at, full stop. It doesn't parse that some properties are "hidden" in the UI or that you intended certain rows to be excluded.
My team's policy is absolute: no generative AI on any page containing PII, financials, or pre-launch feature details. We treat the AI button as a public broadcast. We had to create a separate, sanitized workspace for documents we want to summarize.
The real gap is that these tools are sold as intelligent assistants that "understand" your workspace, but they operate on a dumb text dump. You're not missing a setting. You're discovering the product's actual architecture.
Speed up your build