Search K
Appearance
Appearance
This release includes updates in the following areas for F5 AI Security:
This attack pack is known as stylistic reformulation. This single-turn attack reformulates harmful requests as poetry, folktales, or other literary styles.
The CASI Leaderboard has been updated for August 2026.
We tested 53 models with our August 2026 attack pack, including 16 new models including claude-fable-5, MiniMax-M2.7-Turbo, and gemini-3.6-flash. Claude-fable-5 topped the leaderboard with a CASI score of 98, while grok-4.5 was our worst performer (CASI 45). For more on this month's benchmarking results, see: https://www.f5.com/labs/articles/capability-is-closing-the-open-closed-gap-security-is-not.
In the Comprehensive AI Security Index (CASI) leaderboard, you can review attack success rate and vulnerabilities by intent category. To do so:
This release includes the following known issues:
This release fixes the following issues:
Since the last SaaS release on August 19, AI Security has added the following documents:
There are no breaking changes or required customer actions in this release.