List your company

A1 shortcut attack defense on Real-LLM 8 targets × 3 phrasings full stack late 2025/2026 (val)

0Attack-Induced ISRMockAgent (worst-case bound)
-4235077Jun 17, 2026

Evaluation results

MethodLinks
MockAgent (worst-case bound)ContractGuard level=L3, Temperature=02026.060
Claude Opus 4.8ContractGuard level=L3, Temperature=02026.060
Claude Sonnet 4.6ContractGuard level=L3, Temperature=02026.060
Claude Haiku 4.5ContractGuard level=L3, Temperature=02026.060
GPT-OSS-120BContractGuard level=L3, Temperature=02026.060
Amazon Nova PremierContractGuard level=L3, Temperature=02026.060
Amazon Nova 2 LiteContractGuard level=L3, Temperature=02026.060
MockAgent (worst-case bound)ContractGuard level=L0, Temperature=02026.06100
Claude Opus 4.8ContractGuard level=L0, Temperature=02026.06100
Claude Sonnet 4.6ContractGuard level=L0, Temperature=02026.06100
Claude Haiku 4.5ContractGuard level=L0, Temperature=02026.06100
GPT-OSS-120BContractGuard level=L0, Temperature=02026.06100
Amazon Nova PremierContractGuard level=L0, Temperature=02026.06100
Amazon Nova 2 LiteContractGuard level=L0, Temperature=02026.06100