Guardian agents: New approach could reduce AI hallucinations to below 1%

Vectara has developed an innovative solution called the Hallucination Corrector, which uses guardian agents to detect and correct AI hallucinations more effectively than existing methods. These agents dynamically adjust outputs in agentic workflows, preserving content while ensuring accuracy—e.g., reducing hallucinations below 1% for smaller models. This approach addresses the growing challenge of deploying AI in high-stakes environments, offering a middle ground between full AI integration and avoidance. By leveraging HCMBench, Vectara provides tools to evaluate its system's effectiveness, enabling enterprises to adopt AGI safely while minimizing risks.