AWS clean-rooms: Caveat that metrics noise level is not a privacy guarantee
Summary
Adds a note explaining that the custom metrics noise level only obscures raw values, is not a formal privacy guarantee such as differential privacy, should not be relied on alone to protect sensitive data, and should be combined with controls like limiting metric recipients; also notes providers can review the noise level before allowing table access.
Security assessment
The changed line is explicit data-privacy hardening guidance: it warns that noise is not differential privacy and must be combined with other controls (e.g., restricting metric recipients) to protect sensitive data. It documents a limitation rather than fixing a specific vulnerability.
Evidence
+The noise level helps obscure raw values in custom metrics. It isn't designed to provide a formal privacy guarantee, such as differential privacy. Don't rely on the noise level alone to protect sensitive data. Use it together with other privacy controls, such as limiting which members can receive metrics.
Diff
diff --git a/clean-rooms/latest/userguide/custom-model-logs.md b/clean-rooms/latest/userguide/custom-model-logs.md index 0cd574d..3fd423f 100644 --- a//clean-rooms/latest/userguide/custom-model-logs.md +++ b//clean-rooms/latest/userguide/custom-model-logs.md @@ -24,0 +25,6 @@ When you configure a model algorithm (either via the console or the `CreateConfi +###### Note + +The noise level helps obscure raw values in custom metrics. It isn't designed to provide a formal privacy guarantee, such as differential privacy. Don't rely on the noise level alone to protect sensitive data. Use it together with other privacy controls, such as limiting which members can receive metrics. + +The model author sets the noise level when they associate the configured model algorithm. Training data providers can review the association's privacy configuration, including the noise level, before they allow the association to use their configured table. For more information, see [Associating the configured model algorithm in AWS Clean Rooms ML](https://docs.aws.amazon.com/clean-rooms/latest/userguide/associate-model-algorithm.html). +