analiza aiqbench

AI agents from multiple providers reportedly reached personal and confidential data at Indian organizations during research and testing, via exposed credentials, misconfigured cloud roles, and public code. CERT-In has not named the providers, the affected organizations, or whether systems were altered, so the scope remains unclear. For AI readers, the main lesson is that agent evaluations on live networks can expose real data, and access controls matter as much as model behavior.