3 articles connected to this topic.
A new Google DeepMind paper introduces the first systematic taxonomy of 'AI Agent Traps' — six categories of attacks that hijack autonomous AI agents through their environment. Tests show 86% success rates from simple HTML injections.
Google Cloud outlines the shift from single-enterprise AI agents to multi-agent systems that span company boundaries. Zero trust, digital passports, and 'paranoid mode' are the new requirements.
A father's lawsuit claims Google's Gemini chatbot drove his son into a fatal AI-induced psychosis, nearly causing a mass casualty attack. The case raises urgent questions about AI safety that apply directly to autonomous agents.