Oracle Cloud Infrastructure Documentation / Release Notes

Pin a Guardrails Version in OCI Generative AI

You can now select a specific guardrails version when using the ApplyGuardrails API in OCI Generative AI. Guardrails versions use semantic versioning, such as 1.0.0, to represent the behavior of guardrail protections for content moderation (CM), prompt injection (PI), and personally identifiable information (PII) detection.

This feature gives you more control over guardrail behavior in production. You can pin a specific version to maintain stable behavior, or omit the version configuration to use the default guardrails version.

Use the new ListGuardrailVersions API to review available guardrails versions before selecting one. For each version, the API returns details such as:

  • Guardrails version
  • Lifecycle state, such as active, deprecated, or retired
  • Activation, deprecation, or retirement time, when applicable

To use a specific version, add guardrailVersionConfig to the ApplyGuardrailsDetails request. For example:

"guardrailVersionConfig": {
  "guardrailVersion": "1.0.1"
}

Version 1.0.0 is the initial AI Guardrails release with foundational safety checks for CM, PI, and PII. Version 1.0.1 improves accuracy for CM and PI.

For API examples and setup details, see About OCI Generative AI Guardrails.

Important

Disclaimer

Our Content Moderation (CM) and Prompt Injection (PI) guardrails have been evaluated on a range of multilingual benchmark datasets. However, actual performance may vary depending on the specific languages, domains, data distributions, and usage patterns present in customer-provided data as the content is generated by AI and may contain errors or omissions. Accordingly, it is intended for informational purposes only, should not be considered professional advice and OCI makes no guarantees that identical performance characteristics will be observed in all real-world deployments. The OCI Responsible AI team is continuously improving these models.

Our content moderation capabilities have been evaluated against RTPLX, one of the largest publicly available multilingual benchmarking datasets, covering more than 38 languages. However, these results should be interpreted with appropriate caution as the content is generated by AI and may contain errors or omissions. Multilingual evaluations are inherently bounded by the scope, representativeness, and annotation practices of public datasets, and performance observed on RTPLX may not fully generalize to all real-world contexts, domains, dialects, or usage patterns. Accordingly, the findings are intended to be informational purposes only and should not be considered professional advice.