RESEARCHERS UNCOVER GPT-5 JAILBREAK AND ZERO-CLICK AI AGENT ATTACKS EXPOSING CLOUD AND IOT SYSTEMS

Summary from the TLDR InfoSec newsletter

Researchers from NeuralTrust demonstrated a jailbreak against GPT-5 using "Echo Chamber" and storytelling to bypass ethical guardrails, poisoning context to generate prohibited instructions without triggering refusals. Zenity Labs revealed "AgentFlayer" attacks exploiting ChatGPT Connectors, Cursor AI, and Microsoft Copilot via prompt injections in documents, tickets, or emails to exfiltrate data like API keys. These reveal how AI connected to external systems increases attack surfaces, with Tel-Aviv University showing smart home hijacking via poisoned calendar invites to Google's Gemini AI.