Relative impact of AI agents on cybersecurity offense vs defense

Determine whether AI agents will aid cybersecurity offense more than cybersecurity defense or vice versa, to clarify the net impact of deploying AI-agent capabilities in cybersecurity contexts.

Background

The paper demonstrates that teams of AI agents can autonomously exploit real-world zero-day web vulnerabilities, suggesting significant dual-use implications. While the work shows offensive capabilities, it also notes potential defensive applications, such as assisting penetration testers.

Given these dual-use possibilities, the authors explicitly state uncertainty about whether AI agents will ultimately strengthen attackers or defenders more. Resolving this uncertainty is critical for policy, deployment decisions, and risk management related to AI agents in cybersecurity.