Compiled by the editorial desk with reference to public reviews, expert commentary, and official company announcements.

OpenAI's foray into web browsing with its ChatGPT-powered Atlas browser has met with a lukewarm reception from early adopters, who cite sluggish performance, irrelevant search results, and unresolved security risks. The browser, unveiled earlier this week, is positioned as a step toward a 'super-assistant' that can handle complex tasks like booking flights or purchasing groceries through an 'agent mode.'

Initial reviews from technology journalists paint a picture of a product that, while conceptually ambitious, struggles to deliver a seamless experience. Emma Roth of The Verge noted that ChatGPT's suggestions 'aren't always relevant,' recounting instances where search results included local news that had no connection to her location. The browser's reliance on ChatGPT as a portal to the web, she argued, feels inadequate compared to traditional search engines.

One notable design choice is the prominent link to Google in the top-right corner of every search results page, which Roth interpreted as an implicit acknowledgment of the browser's limited search capabilities. This fallback to a familiar search engine suggests OpenAI is aware of the current constraints of its AI-driven approach.

Agent Mode Underwhelms in Speed and Accuracy

The flagship 'agent mode,' designed to automate multi-step tasks, has been a particular point of criticism. In a test by Roth, the browser took ten minutes to add three items to an Amazon cart, a task that Perplexity's Comet completed in two minutes. Wall Street Journal columnist Nicole Nguyen observed that Atlas 'struggled with clicking the correct button,' comparing its performance to a toddler eating—inefficient but ultimately successful. In one instance, finding flights for an upcoming trip consumed 16 minutes.

These early tests suggest that while the agent mode can eventually complete tasks, its speed and reliability are far from polished. The browser's underlying architecture, built on Google's Chromium, places it in the same category as other AI-enhanced browsers, but reviewers note that its functionality currently mirrors existing offerings without adding significant value.

Security Experts Raise Red Flags

Beyond usability, cybersecurity concerns have emerged as a significant hurdle. Brave, a competing browser company, recently outlined vulnerabilities in Perplexity's Comet, particularly around 'prompt injections,' where hidden instructions can manipulate AI systems into harmful actions. George Chalhoub, an assistant professor at UCL Interaction Center, told Fortune that such risks are inherent to systems that interpret natural language and execute actions. He warned that an AI agent in a browser could become 'a potential attack vector,' potentially extracting emails, passwords, or personal data from connected accounts.

OpenAI has not yet detailed how it plans to address these security challenges for Atlas. The company's broader ambitions, however, extend beyond the browser itself. Earlier this month, OpenAI introduced an Apps SDK, allowing developers to build applications within ChatGPT, signaling a move toward a more integrated operating system.

As OpenAI prepares to invest over $1 trillion in data center infrastructure over the next year, the early reception of Atlas raises questions about the readiness of its AI products. While the browser is an initial step, the gap between its current performance and the promised 'super-assistant' experience remains wide. For now, users may find that traditional browsers, augmented by AI features, offer a more reliable and secure browsing experience.