ChatGPT’s Browser Bot Appears to Block New York Times Links
▼ Summary
– AI-powered browsers like ChatGPT Atlas can perform tasks autonomously but avoid certain sources, especially those whose parent companies are suing OpenAI.
– These browsers operate in “agent mode” by mimicking human users, appearing as normal Chrome sessions to bypass restrictions on automated crawling.
– Unlike standard web crawlers that respect blocking instructions, agentic browsers can access pages that typically prohibit automated access.
– When asked to summarize articles from litigious publishers, Atlas circumvented direct access by gathering information from alternative sources like social media and other news outlets.
– Atlas preferentially used AI-friendly sources with agreements with OpenAI, avoiding direct interaction with companies involved in legal disputes.
AI-powered browsers such as ChatGPT Atlas offer more than just a sidebar chatbot, they possess what developers call “agentic capabilities,” enabling them to perform complex tasks like booking flights or hotels. However, recent findings suggest these intelligent agents may deliberately avoid accessing content from certain publishers, particularly those engaged in legal disputes with their parent companies.
When operating in agent mode, Atlas navigates the web on behalf of the user, appearing in site logs as a regular Chrome session since it’s built on Chromium. This allows it to bypass restrictions that typically block automated crawlers. While this design supports seamless user interaction, it also raises questions about how the agent handles content from organizations that are suing OpenAI.
An investigation revealed that when asked to summarize articles from The New York Times or PCMag, both of which have parent companies involved in copyright litigation against OpenAI, the agent took elaborate detours rather than accessing the original sources directly. For PCMag, it gathered information from social media posts and other news sites that referenced the article. For The New York Times, it pieced together a summary using reporting from outlets like The Guardian, The Washington Post, Reuters, and the Associated Press. Notably, all but Reuters have existing content or search agreements with OpenAI.
This behavior suggests the agent is programmed to avoid litigious publishers, opting instead for alternative sources that present a lower legal risk. Rather than retrieving information directly, it constructs summaries from secondhand reports, effectively sidestepping potential conflicts. This strategy highlights the delicate balance between providing useful assistance and minimizing corporate liability in an increasingly contentious digital environment.
(Source: Gizmodo)
