Researchers Uncover GPT-5 Jailbreak and Zero-Click AI Agent Attacks Exposing Cloud and IoT Systems
Researchers studying cybersecurity have discovered a jailbreak method that allows them to generate illegal instructions by evading ethical restrictions put in place by OpenAI in its most recent large language model (LLM), GPT-5.
NeuralTrust, a generative artificial intelligence (AI) security platform, claimed to have used narrative-driven steering in conjunction with a well-known method known as Echo Chamber to fool the model into generating unwanted results.
Security researcher Martí Jordà explained, "We use Echo Chamber to seed and reinforce a subtly poisonous conversational context, and then we guide the model with low-salience storytelling that avoids explicit intent signaling." This combination reduces triggerable refusal indications read more about Researchers Uncover GPT-5 Ja...