Adversa AI evaluated 100 AI agents across ten categories and found that only 11% are both capable and well-defended, with most agents exposed to the “lethal trifecta” of private data access, untrusted content exposure, and outbound actions. The report warns that computer agents and coding agents have the largest attack surfaces and blast radii, making careful controls over egress, identity, and irreversible actions essential. #AdversaAI #AIRiskQuadrant #computeragents #codingagents
Keypoints
- Adversa AI tested 100 agents across ten categories.
- Only 11% were classified as capable and well-defended.
- Most agents had the “lethal trifecta” of risk.
- Computer agents can expose an entire machine if compromised.
- Coding agents increase supply-chain risk through broad tool access.
Read More: https://www.securityweek.com/security-of-100-ai-agents-tested-and-ranked-what-you-need-to-know/