AI's Specialized Surge: Cybersecurity Models, Benchmark Dominance, and the Safety Tax

Specialized AI models are outperforming generalist models in high-stakes domains, but the cost and safety trade-offs are becoming steeper.

Evidence For

• Google's Gemini 3.8 Flash Cyber achieved 86.2% on the CyberGym benchmark and 47.2% on CWE-Bench, matching frontier-level performance in autonomous vulnerability discovery and patching at scale.
• Anthropic's Claude Fable 5.1 scored 66 on the Artificial Analysis Intelligence Index, pulling six points clear of...