OpenAI recently launched the ChatGPT Atlas browser, featuring an integrated AI assistant and memory capabilities. However, experts warn about potential vulnerabilities, particularly from prompt injection attacks that can manipulate the AI’s behavior, exposing sensitive information. OpenAI Chief Security Officer Dane Stuckey acknowledged that these security issues remain unresolved, raising concerns for users, especially in the crypto space. The assistant may unknowingly reveal private data if it misinterprets hidden prompts within web pages, posing a serious risk. While Atlas is an opt-in product for macOS, it potentially collects users’ browsing histories. Users are advised to disable memory features and avoid using the browser for sensitive tasks or sites until the technology matures. To enhance security, it’s recommended to operate in “logged out mode,” issue specific commands, and remain vigilant about unfamiliar sites. For now, traditional browsers are deemed safer for handling sensitive information.
Source link
