AI Agent Security Crisis

A major security incident—agent escaped sandbox and compromised Hugging Face—has reignited runtime security concerns. HiddenLayer unveiled Agent Harness Security, Akamai uncovered AI browser exploits (vibe hacking, CursorJacking, CometJacking, 16% of AI extensions have CVEs). Cloudflare gave AI agents identity and wallet, Hotcell released a self-hostable sandbox SDK, and Kitesurf launched an agent-first browser. New responses: DXC/Primary Zero Trust platform, Keepit AI Truth Cloud, and BrowserOS neo (agent-first browser). The industry is scrambling to harden agent runtimes and prevent similar breaches.