Privacy and security experts sound alarm over OpenAI's new browser

OpenAI's recently released Atlas browser for macOS has raised significant privacy and security concerns due to its advanced artificial intelligence (AI) capabilities. Atlas, which features an integrated ChatGPT agent, is designed to autonomously complete tasks on websites and remember user activity from previous searches, a practice that requires it to collect considerably more sensitive data than traditional browsers.

Privacy advocates are concerned about the extent of data collection, with one technologist reporting that Atlas memorised highly sensitive health queries, including a real doctor's name. Meanwhile, security researchers have flagged serious vulnerabilities related to prompt injection attacks, in which malicious commands hidden in websites could trick the autonomous agent into violating its rules, potentially leading it to book hotels, send messages, or delete files. Researchers have already demonstrated how Atlas can be tricked into visiting malicious sites.

OpenAI has stated that Atlas is not intended to retain sensitive information and has implemented controls to prevent agents from running code or using autofill data. However, the company's CISO admitted that prompt injection attacks remain a largely "unsolved security problem" across all AI platforms. Security experts caution that the risks associated with AI browsers may currently outweigh their benefits.

Training Announcement: Freevacy offers a one-day course on AI Literacy, providing a foundational understanding of AI tools and applications, along with their practical applications in the workplace. The interactive session covers basic AI concepts, ethical considerations, and hands-on demonstrations of AI tools. Attendees will acquire the knowledge and confidence to effectively integrate AI solutions into their daily tasks while ensuring their responsible use. No prior experience in AI is required. Find out more.