Compare
Azure AI Content Safety vs Guard
Azure AI Content Safety filters harmful content. Guard produces Australian regulatory evidence. Content safety is not compliance.
Why this comes up
Azure AI Content Safety is built to detect harmful content — violence, hate speech, self-harm, and sexual material — across text and images, with severity scoring. For trust-and-safety use cases it’s a strong, well-tuned service.
But harmful-content detection and regulatory compliance are different jobs. Azure’s PII handling is US-focused and generic; it has no Tax File Number, Medicare, or ABN detection with the checksum validation those identifiers need. And it produces no per-call cryptographic attestation and no mapping to APRA CPS 234, the Privacy Act, or ADM transparency.
A regulated Australian business can run every prompt through Azure Content Safety and still have no answer when APRA asks "prove your cross-border PII control was active on this call." That question is about evidence, not toxicity. Guard is built to answer it.
Side by side
| Capability | Azure AI Content Safety | 40° South Guard |
|---|---|---|
| Harmful-content detection (violence, self-harm, hate) | ✓ | ✗ |
| Image content moderation | ✓ | ✗ |
| Australian PII (TFN, Medicare, ABN) with checksums | ✗ | ✓ |
| Per-call cryptographically signed attestation | ✗ | ✓ |
| CPS 234 / Privacy Act regulatory mapping | ✗ | ✓ |
| APP 8 cross-border disclosure flagging | ✗ | ✓ |
| 7-year tamper-evident audit trail | ✗ | ✓ |
✓ = supported · ~ = partial · ✗ = not supported
Download the full comparison (PDF)Could you run them together?
Yes. If harmful-content filtering matters to your use case, run Azure Content Safety for that and Guard for the compliance evidence layer.
They don’t overlap: one is about whether output is toxic, the other is about whether your regulatory controls ran and can be proven.
See Guard on your own AI calls
Book a demo and we’ll show you a signed attestation for a real call — mapped to your obligations under CPS 234, the Privacy Act, and ADM transparency.