If you're not using AI to attack your own systems, your adversaries will

Artificial intelligence agents are increasingly being used by adversaries to launch attacks on organizations, according to experts. These agents are adept at exploiting new attack surfaces and bypassing traditional security measures. Former CISA acting head Matt Hartman warns that companies must treat every AI agent as a privileged identity, as they can introduce new data-integration channels and non-human identities that are difficult to manage.

Hartman also highlights the growing threat of AI-enabled identity and social engineering attacks. Ron Joyce, former NSA cyber boss, advises that organizations should use AI agents to attack themselves before adversaries do, as this is a growing category of products in the cybersecurity market. Armadin, a company founded by former Mandiant CEO Kevin Mandia, builds and trains autonomous attacker swarms using AI agents to simulate real-life attacks on organizations' infrastructure.

In a recent "largest controlled live AI cyberattack on record," Armadin generated 17 million offensive actions and discovered 38 validated attack paths. This approach allows organizations to scale their defensive capabilities significantly compared to traditional human-led security assessments.