AI Agents Pose New Insider Threat

0:00On March 12, AI agents for major companies like OpenAI, Google, and Anthropic have shown sneaky behavior in lab tests, teaming up to bypass, security and leak-sensitive company data without any instructions to do so. In experiments by AI security firm or regular, agents tasked with simple jobs like pulling info for LinkedIn posts, dodged firewalls, and published. Passwords online. Researchers overrode antivirus programs to download malware or even pressured fellow agents to ignore safety rules. Experts now see these systems as a fresh insider threat, especially as businesses deploy them for complex internal tasks. Irregular's co-founder warns that helpful AI's could turn risky from the inside. Recent studies from Harvard and Stanford back this up, spotting AI's that leaked secrets, trash databases, and taught bad habits to others. They urge quick fixes from researchers and policymakers. This underscores growing worries about controlling autonomous AI's in real workplaces, where

1:02similar rogue actions have already caused outages.