The weaponization of agentic AI is shifting cyberattacks from AI-assisted to fully autonomous, with defenders using AI agent swarms to identify vulnerabilities and strengthen protection, creating a machine-versus-machine security landscape. Frontier AI now represents a new attack vector with evolving security challenges, as highlighted by Anthropic’s Mythos early preview model and Project Glasswing, which raise awareness of undetected vulnerabilities. Traditional cybersecurity approaches assume defenders have time to respond, but frontier AI reduces the mean time to discovery of infrastructure vulnerabilities and can accelerate data exfiltration from weeks and days to hours and minutes. Attackers are evolving from AI-assisted to AI-automated campaigns, requiring defenders to do the same for detection, response, governance, and remediation. Major cybersecurity infrastructure providers, including Cisco, Hewlett Packard Enterprise, Microsoft, and Palo Alto Networks, are investing in runtime AI security through acquisitions and organic roadmap development. The number of new AI-infused tools is overwhelming, spanning identity, network, data, and cloud protection, so frontier AI requires a new cybersecurity playbook. Both established cybersecurity incumbents and a new wave of AI security infrastructure start-ups are addressing the challenge of securing modern AI.