{"id":344618,"date":"2025-08-11T01:30:00","date_gmt":"2025-08-11T08:30:00","guid":{"rendered":"https:\/\/cms-articles.softonic.io\/es\/?p=392776"},"modified":"2025-08-11T01:30:26","modified_gmt":"2025-08-11T08:30:26","slug":"chatgpt-has-a-serious-security-problem-but-users-are-not-aware-of-it","status":"publish","type":"post","link":"https:\/\/cms-articles.softonic.io\/en\/chatgpt-has-a-serious-security-problem-but-users-are-not-aware-of-it\/","title":{"rendered":"ChatGPT has a serious security problem, but users are not aware of it"},"content":{"rendered":"\n

Researchers have revealed a “jailbreak” technique that evades the ethical restrictions imposed by OpenAI<\/a> on its GPT-5 language model, using an approach called Echo Chamber<\/strong>. This technique, combined with contextual narratives, allows users to make queries that would normally be rejected by the model, facilitating the generation of undesirable responses. According to Mart\u00ed Jord\u00e0, a cybersecurity researcher, this method is based on introducing a subtly toxic conversation context that does not emit direct signals of malicious intent.<\/p>\n\n\n

Beware, AI fans<\/h2>\n\n\n

These potential attacks, which are framed within a “persuasion loop,” pose an increasing risk as generative language models are used in business environments<\/strong>. Recent findings have shown that attackers may choose keywords and construct phrases that lead the model to reveal dangerous instructions, such as in the case of creating Molotov cocktails, in a narrative format that masks the direct request.<\/p>\n\n\n

Additionally, new attacks<\/a> known as ‘zero-click’ have been identified, where confidential information can be extracted from seemingly harmless documents and emails<\/strong> through prompt injections. These attacks exploit the integration of AI models with external systems, further exposing security vulnerabilities.<\/p>\n\n\n

\n