May 21, 2026 - SaaS - v10.31.0

This release includes updates in the following areas for F5 AI Security:

• AI Guardrails
• AI Red Team
• Platform
• Resolved issues

---

AI Guardrails

Improved guardrail performance

The AI model that powers AI Guardrails has been updated. Tests show the following improvements to AI Guardrails performance, with up to:

• 1.6x greater throughput
• 9x increased context window size
• 4% higher detection accuracy

Update your AI Guardrails

Because of the model update, all GenAI Guardrails need to be updated. How you update depends on whether you use pre-defined F5 AI Guardrails or custom guardrails.

• Custom regex and keyword guardrails are not affected.

F5 AI Guardrails

If you've used F5 AI Guardrails before now, read the following F5 KB article: Update guardrails.

Custom guardrails

If you use custom guardrails, you need to update only the GenAI scanners. When a new version is available, it appears in your version history. You can publish it when you're ready. However, it won't become active until you set it as the active version.

This improvement does not change hardware requirements.

AI Red Team

Evaluator improvements

The refusal evaluator is now more accurate. Improvements include:

• More specific prompting based on attack type
• A better version of few-shot refusal checks

Based on internal testing, these changes deliver a 50–60% reduction in false positives.

In-progress campaigns now link to reports

Under Red Team > Reports, in-progress campaigns now include a direct link to the campaign report.

Platform

New authentication options for connections

You can now connect to model providers that require OAuth 2.0 or AWS authentication. To set this up, go to Configure > Connections > Other models > Authentication and select the authentication type of your choice.

Improved custom role support

Custom role creation includes support for creating a true read-only role. If you invite someone with the default User role, they can access the chat feature but can no longer create personal API tokens.

For more information, see Manage users and roles.

Permissions reorganized

When you create or manage roles, you'll now find the API and Branding permissions in the Organization section.

Admins can set the default role

Admins can now select the default role assigned to new users. Hover over the desired default role and select Set as default.

---

Resolved issues

This release fixes the following issues:

Custom roles

• Custom role name validation prevented you from saving your work.

AI Guardrails

• Block outcomes weren't reflected correctly in the logs view when flag_only was set to false.
• A Comprehensive AI Security Index (CASI) score of 0 was displayed as N/A.
• Adding or removing a guardrails package didn't work in the UI.
• The progress bar showed 100% before attacks were complete.
• The Save button wasn't visible when adding a user.
• The Auto-remediate checkbox didn't explain why it was unavailable in certain states. A tooltip has been added.
• A translation key was missing in the downgrade AI Guardrails dialog.