Emerging AI-powered browsers like ChatGPT Atlas and Perplexity’s Comet are reportedly capable of bypassing digital paywalls and content blockers. According to a recent investigation, these AI tools were able to reconstruct paywalled articles when prompted. While initial tests by this publication could not replicate the specific bypass, if confirmed, such capabilities could pose a significant threat to news platforms and other content sites that rely on subscriptions for premium content.
Concerns Rise Over AI Browsers Accessing Unauthorized Content
The Columbia Journalism Review highlighted instances where both ChatGPT Atlas and the Comet browser successfully generated articles that were protected by paywalls. Although other AI browsers like Edge’s Copilot mode and The Browser Company’s Dia exist, this behavior was most prominent in Atlas and Comet. Both browsers are publicly accessible, with Comet notably providing advanced ‘agentic’ functionalities to all users.
It’s worth noting that when our own tests were conducted using the Comet browser with similar prompts on the same websites mentioned by the CJR, the browser declined to provide the paywalled content. This discrepancy could be due to recent updates by the browser developers to modify the AI agent’s behavior, or perhaps the original investigation employed more sophisticated prompt injection techniques not detailed in their findings.
The report further detailed that Atlas and Comet successfully retrieved the complete text of a lengthy article (approximately nine thousand words) from the MIT Technology Review, which was protected by a paywall. By simply asking the AI browsers to “print the text of this article,” users were allegedly able to view the full content without needing a paid subscription. Interestingly, using the same prompt in a standalone chatbot interface did not yield the same outcome.
As suggested by the report, the core reason for these alleged unauthorized actions lies in how the underlying AI agents operate. They appear “indistinguishable from a person” using a conventional browser. Unlike traditional crawlers used by search engines like Google Chrome or Mozilla Firefox, which follow strict protocols for data scraping and display, these AI agents behave differently.
When these AI agents visit a website, they reportedly employ a digital signature that mimics a human user, interacting with the site just as a person would. This makes it challenging for websites to identify and block these ‘crawlers’ or to flag their activities as illegitimate. Even if websites attempt to block them using standard protocols like Robots Exclusion Protocol, these agents can reportedly modify their digital signature to evade detection.
Beyond paywall circumvention, the report points to a broader concern: AI browsers can be commanded to visit websites and summarize articles. This implies users could access and digest content without actually visiting the website directly, thus depriving publishers of crucial ad revenue generated by page views.
While chatbots already possess summarization abilities, embedding these functions directly into browsers could accelerate the decline in search traffic for content-driven websites. The long-term impact will undoubtedly hinge on the wider adoption of AI browsers and the forthcoming data privacy regulations from policymakers.