Artificial intelligence agents pose significant security concerns, particularly as their capabilities expand and they become more integrated into various systems. The threat model for these agents is shifting from criminal to geopolitical, driven in part by state-aligned activity involving major technology companies like Intel. This change necessitates a different approach to security, one that accounts for the unique risks and motivations of nation-state actors. Researchers at Perplexity, drawing on their experience with general-purpose agentic systems, have outlined key considerations for securing AI agents, including the need for robust testing and validation protocols1. As AI agents become increasingly ubiquitous, their security will depend on the ability to adapt to evolving threats and mitigate potential vulnerabilities. The security of AI agents matters to practitioners because it requires a proactive and nuanced approach to threat modeling, one that acknowledges the complex interplay between technological, social, and geopolitical factors.