Anthropic dominates Cisco's LLM Security Leaderboard
Eight of the top ten spots are held by Anthropic's models, with Claude Opus 4.5 in first place, followed by Sonnet 4.5 and Haiku 4.5. Cisco evaluated models' resilience to multi-step attacks simulating real hacker behavior.
OpenAI's GPT-5.2 and GPT 5 Nano placed seventh and ninth. Models from Mistral, DeepSeek, Cohere, and xAI ranked lower. Cisco notes that over 80% of companies plan to deploy AI agents, but only a third are prepared to protect enterprise data.
Related stories
- Fable 5.1 refuses the knife but heats a gas can anyway
- Opus 5 falls to prompt injection 2% of the time
- Claude Opus 5 broke 11 truces and won Vending-Bench with $11,182
- Anthropic's models miss the frontier in a security PR-review test
- Semgrep: GLM-5.2 outperforms Claude Code in IDOR detection
- Safety research on Fable 5 and Opus 4.8 models
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.