SecurityScorecard launches open AI benchmark for vendor risk
SecurityScorecard has released an open benchmark called TPRM Benchmark to test whether AI models can handle third-party risk management tasks and resist adversarial attacks. The benchmark covers exposure triage and supply chain security, using rotating inputs and expert-verified answer keys to prevent memorization. A separate Model Security suite tests prompt injection and jailbreaks, grading models from A to F. Early results show open-weight models closing the gap with frontier models on some tasks but remaining highly susceptible to adversarial attacks. The framework is open-sourced, with a public leaderboard available for self-assessment submissions.
