🤖 Real AI Agent Security Test: LLM Spotted the Hack, Tools Ignored It Everyone figured modern LLMs had security licked. Then agent-probe hit a real AI agent—and exposed a killer flaw in the tool layer. 4 min read 1 month ago