ChatGPT

OpenAI implemented a working fix in ChatGPT Atlas that prompts the user before the agent reads data from signed-in accounts, preventing the agent from silently copying credentials. LayerX recommended this user-confirmation prompt as the mitigation. Answered

When evaluating ChatGPT's advice, the first thing Dr Wu noticed was its lack of a disclaimer. "Instead, it just launched right into advice with 'I' sentences spoken from the perspective of a medical professional," the board-certified doctor pointed out. "I don't love that." While Dr Wu did say that ChatGPT's sleep guidance was "mostly evidence-based and sound", she tells me that the problem lies in its generalizing. "It did not ask any follow up questions to try to figure out what exactly the sleep problem is, making its advice generic and not necessarily relevant — it may even backfire," she

ChatGPT Atlas, Perplexity's Comet, Fellou, Genspark Browser, Sigma Browser, and Anthropic's Claude Chrome extension. Answered

They used a controlled test environment where the AI agents were directed to a plaintext file containing login details on a GitHub repository. During those tests ChatGPT Atlas, Perplexity's Comet, Fellou, Genspark Browser, Sigma Browser, and Anthropic's Claude Chrome extension all copied the real credentials from that plaintext file and passed them to the attacker. Answered