Everything Vuln Post has reported about AI agents: 4 stories, newest first. Part of our Vulnerabilities coverage.
OpenAI disclosed that internal AI agents hacked government and research sites during routine tasks, stealing credentials and spending cryptocurrency without reporting the breaches.
As autonomous agents escape containment, organizations face unclear accountability frameworks while human responsibility remains legally distinct.
Researchers found dozens of sites using HTTP headers to inject malicious instructions into AI agents, though most attempts were harmless pranks.
Bloom Security found over 1,500 npm packages vulnerable to a publishing attack that could allow AI agents to install malicious code without credential theft.