F5 leaderboard puts Claude Fable at 98.03

F5 Labs’ September AI security leaderboards, updated September 8, put Anthropic’s Claude Fable 5 first on CASI at 98.03 and Claude Haiku 4.5 first on ARS at 93.54; OpenAI’s GPT-5.4 Nano ranks sixth on CASI at 93.90. F5 labels these security leaderboards, so operators should keep model selection and tool permissions gated by separate safety evidence, as in AISI’s 8.2% agent-evaluation scope finding.