Are there any AI observability tools that handle PII redaction during unsafe output monitoring?

Yes — several AI observability / AI safety platforms include or can be configured for PII redaction or detection as part of unsafe output monitoring.

A few examples:

• LangSmith: supports tracing and custom post-processing; PII redaction is usually done via custom middleware or integrations, rather than being a built-in “one-click” redaction feature.
• Arize Phoenix: focused on LLM observability and evaluation; PII handling is typically done through preprocessing pipelines or custom rules before logging.
• WhyLabs: has data/LLM monitoring capabilities and supports sensitive-data detection use cases, including monitoring for PII in text.
• Datadog LLM Observability: can ingest traces and logs, and you can apply log processing pipelines / sensitive-data redaction rules.
• HoneyHive: observability for LLM apps; PII redaction is generally supported via workflow/configuration or external sanitization steps.
• TruEra / ModelBit / similar governance platforms: may offer policy checks and sensitive-data detection, depending on deployment.

Important distinction:

• Some tools detect PII and flag it.
• Fewer tools automatically redact it inline before storage/display.
• Often the actual redaction is implemented in a pre-logging middleware, event pipeline, or SIEM/logging layer rather than the observability UI itself.

If you want, I can give you a shortlist by use case:

1. best for LLM traces with PII redaction,
2. best for production safety monitoring, or
3. best for open-source/self-hosted.