PRODUCTION SCENARIO
A children's tutoring app on Gemini must block responses that carry even a low likelihood of harassment or dangerous content. A separate internal engineering assistant finds those same settings too restrictive when it discusses security topics.
How should the platform team handle the two applications?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Set the blocking threshold per harm category separately for each app
Gemini's configurable content filters block by harm category at a threshold the customer sets, and Google says to choose it for what is appropriate to each use case and business, so the two applications get different settings. System instructions steer the model but sit outside the filters, the non-configurable CSAM and personal-data filters can never be switched off, and one company-wide threshold recreates the conflict.