Question 210
An audit of a Cortex-enabled workflow finds the following issue: A support chatbot sees a customer’s authentication token in an attached log. What evidence would most reliably confirm production readiness for this requirement?
Explanation
Correct Answer
“Verified exclusion of unnecessary sensitive fields, role-compliant access and safe logging/retention of inference data” is supported by this requirement: A support chatbot sees a customer’s authentication token in an attached log. Data minimization and access controls should precede model processing and telemetry storage. Verify it using verified exclusion of unnecessary sensitive fields, role-compliant access and safe logging/retention of inference data.
Incorrect Answers
“Total prompt length as evidence that privacy is protected” would not meet the requirement that a support chatbot sees a customer’s authentication token in an attached log. Data minimization and access controls should precede model processing and telemetry storage. The more reliable approach is to classify sensitive fields, limit prompt input to the necessary authorized data and verify downstream logs do not expose restricted values.
“Frequency of secrets appearing in model responses” would not meet the requirement that a support chatbot sees a customer’s authentication token in an attached log. Data minimization and access controls should precede model processing and telemetry storage. The more reliable approach is to classify sensitive fields, limit prompt input to the necessary authorized data and verify downstream logs do not expose restricted values.
“Number of unrestricted log copies retained by default” would not meet the requirement that a support chatbot sees a customer’s authentication token in an attached log. Data minimization and access controls should precede model processing and telemetry storage. The more reliable approach is to classify sensitive fields, limit prompt input to the necessary authorized data and verify downstream logs do not expose restricted values.