Redazione RHC : 5 November 2025 16:22
How can publishers protect themselves from AI-powered “smart” browsers if they look like ordinary users? The emergence of new AI-powered “smart” browsers is challenging traditional methods of protecting online content.
OpenAI’s recently released Atlas browser, as well as Perplexity’s Comet and Microsoft Edge’s Copilot mode, are becoming tools that can do much more than display web pages : they perform multi-step tasks, such as gathering calendar information and generating news-based client briefings.
Their capabilities are already posing serious challenges to publishers seeking to limit the use of artificial intelligence in their content. The problem is that these browsers are outwardly indistinguishable from regular users.
When Atlas or Comet access a site, they are identified as standard Chrome sessions, not automated crawlers. This makes them impossible to block using the bots exclusion protocol, since blocking such requests could simultaneously prevent access by regular users. TollBit ‘s ” State of the Bots ” report notes that the new generation of AI visitors is “increasingly human-like,” making monitoring and filtering such agents more challenging.
Another advantage for AI-powered browsers is the way modern paid subscriptions are structured. Many websites, including MIT Technology Review, National Geographic, and the Philadelphia Inquirer , use a client-side approach: the article is loaded in full but hidden behind a pop-up window offering a subscription . While the text remains invisible to humans, it is accessible to AI. Only server-side paywalls, like those of Bloomberg or the Wall Street Journal, reliably hide content until the user logs in. However, if the user is logged in, the AI agent can freely read the article on their behalf.
OpenAI Atlas has received the full text of a subscriber-exclusive article from MIT Technology Review (CJR).
In testing, Atlas and Comet easily extracted the full text of classified MIT Technology Review publications , despite the restrictions imposed by enterprise crawlers like OpenAI and Perplexity.
In one instance, Atlas even managed to reassemble a blocked PCMag article by combining information from other sources, such as tweets, aggregators, and third-party citations . This technique, dubbed “digital breadcrumbs,” was previously described by online research specialist Henk van Ess.
OpenAI says that content viewed by users through Atlas isn’t used to train models unless the “Browser Memories” feature is enabled. However, ” ChatGPT will remember key details about pages viewed,” which, as Washington Post columnist Jeffrey Fowler noted, makes OpenAI’s privacy policy confusing and inconsistent. It’s still unclear to what extent the company uses data obtained through paid content.
A decidedly selective approach is observed: Atlas avoids directly contacting websites that have filed lawsuits against OpenAI , such as the New York Times, but still tries to circumvent the ban by compiling summaries of the topic from other publications —The Guardian, Reuters, the Associated Press, and the Washington Post—that have licensing agreements with OpenAI. Comet, by contrast, shows no such restraint.
This strategy turns the artificial agent into an intermediary that decides which sources are considered “acceptable.” Even if the publisher manages to block direct access, the agent simply replaces the original with an alternative version of events. This alters the very perception of the information: the user receives not an article, but an automatically generated interpretation.
AI-powered browsers haven’t yet achieved widespread adoption, but it’s already clear that traditional barriers like paywalls and crawler blocking are no longer effective. If these agents were to become the primary means of reading news, publishers will need to find new mechanisms to ensure transparency and control over how their content is used by AI.
Redazione