analiza aiqbench

The study reveals that human oversight in AI coding agents is insufficient, with operators missing approximately one-third of high-risk requests involving sensitive infrastructure. This highlights a critical gap between current safety guardrails and the actual threat landscape posed by autonomous code generation tools. For developers deploying these agents, the findings underscore the urgent need for more robust automated detection mechanisms rather than relying solely on human review.