hello,
I activated the centralized guardrails to mask the PII informations,
how can I test this guardrail ?
how can I be sure that the informations are masked?
try sending any sensitive info and check the trace logs of the llm and see what data is sent
cheers
Hi @Sal23,
Try developing an agent which uses any of the LLM and test it by intentionally sending known PII and then verify what reaches the model and what is returned.
If the raw PII never leaves the AI Trust Layer and the prompt contains only masked values, then the guardrail is functioning as expected.
Use the AI Trust Layer’s audit logs or tracing capabilities to confirm that the PII masking guardrail was applied before the request was sent to the LLM model.
Cheers!
I already have a conversational agent that use like a tool an autonomous agent,
the query to the conversational agent is : which client has the mail : ahmedhassan@example.com (which is a PII and It should be masked)
but in the photo (audit), I can find the mail in the logs of the call of this tool, and also the last sentence : no PII/PHI mask
the PII Mask is activated, if not, it should stop the execution (according to the documention of PII mask)
In the policy did you select email entity to be masked?
Also if pii is detected then its masked and it does not fail
Cheers
