Summary: AI-powered browsers like ChatGPT Atlas don’t just fetch information—they actively navigate the web to complete tasks. However, when it comes to certain news sources like The New York Times and PCMag, which are involved in lawsuits against OpenAI, Atlas seems to avoid direct access. Instead, it cleverly gathers information from alternative, safer outlets, reflecting a cautious approach to content from companies currently in legal disputes with its parent company.
Agentic Browsers: More Than Just ChatGPT on the Side
AI-powered browsers such as ChatGPT Atlas aren’t your typical web browsers with a small ChatGPT window answering questions. They come equipped with agentic capabilities, meaning they can perform tasks on your behalf—like booking airline tickets or making hotel reservations. Though Atlas hasn’t yet earned top marks as a travel agent, its ability to autonomously browse and act online is impressive.
When Browsers Sense Danger: Avoiding Litigious Sources
But what happens when the browser’s little web-crawling bot encounters potential risks? The danger here isn’t to the user, but to the browser’s parent company. An investigation by Aisvarya Chandrasekar and Klaudia Jaświńska of the Columbia Journalism Review revealed that when Atlas operates in agent mode—crawling the internet to gather information—it deliberately avoids certain sources. Notably, these sources are owned by companies currently suing OpenAI over copyright issues.
How Atlas Navigates Around Restricted Content
Traditional web crawlers are designed to respect website rules, avoiding pages that disallow crawling. If you use the ChatGPT app to extract information from such pages, it typically responds that it can’t access that content. However, agentic browser modes like Atlas pretend to be a regular user, appearing in site logs as normal Chrome sessions because Atlas is built on the open-source Chromium browser. This allows it to access pages that usually block automated bots, ensuring you can manually browse those sites within Atlas.
The Case of PCMag and The New York Times
Chandrasekar and Jaświńska tested Atlas by asking it to summarize articles from PCMag and The New York Times—both owned by companies involved in active litigation with OpenAI. Instead of directly accessing these sites, Atlas took a roundabout route, like a cautious rat navigating a maze with electrified traps.
For PCMag, Atlas scoured social media and other news outlets, finding citations and tweets containing parts of the article. Regarding The New York Times, it generated summaries based on reporting from alternative sources such as The Guardian, The Washington Post, Reuters, and the Associated Press. Interestingly, all except Reuters have content or search agreements with OpenAI.
What This Means for AI Browsing and Content Access
In both cases, Atlas clearly preferred safer, AI-friendly sources over directly engaging with litigious publications. This behavior highlights the complex interplay between AI browsing capabilities and ongoing legal challenges, suggesting that AI tools may self-regulate content access to avoid potential conflicts.