analiza aiqbench

This report marks a critical inflection point where generative AI is no longer just an efficiency tool but a primary engine for automating complex cyberattacks like LDAP pass-back and SQL exfiltration. The attacker's reliance on older model versions highlights how safety guardrails in frontier models can inadvertently create vulnerabilities by forcing adversaries to downgrade capabilities or exploit less-restricted variants. For the industry, this underscores that AI-driven threat actors now possess autonomous reasoning skills previously reserved for elite state-sponsored groups, demanding immediate updates to defensive strategies and governance frameworks around LLM usage.