A security researcher removed safety guardrails from an open-source AI model and tasked it with identifying vulnerabilities in household devices and networks. The AI successfully found and exploited weaknesses in multiple systems, including a personal computer, but also provided clear guidance on how to remediate the issues.

This experiment demonstrates that AI agents can perform both offensive and defensive security work when properly directed. The dual-use nature—finding vulnerabilities while also helping fix them—highlights both the capability and the governance challenges.

What This Means for Your Business

Security teams can explore AI agents as partners in penetration testing and vulnerability discovery. The technology is ready for internal security assessments. However, the ease with which safety constraints can be removed underscores the need for strict access controls around powerful AI models. Organizations considering AI-powered security tools should invest in both the tools and the governance frameworks to prevent misuse.