The s1ngularity attack: When attackers prompt your AI agents to do their bidding

In mid-September, attackers pushed malicious updates to popular npm packages that installed a post-install script (telemetry.js) which detected local LLM CLI agents (Claude, Gemini, AWS Q) and used crafted prompts to recursively search hosts for sensitive files and steal developer credentials, representing a novel escalation by weaponizing local AI agents to bypass guardrails and automate secret discovery.