ChatGPT’s Browser Bot Seems to Avoid New York Times Links Like a Rat Who Got Electrocuted

AI-powered browsers like ChatGPT Atlas aren’t the only few browsers ChatGPT picture-in-picture box Answers to questions are off to the side. They also have “agentic powers,” meaning they can theoretically perform tasks such as buying airline tickets and making hotel reservations. (Atlas doesn’t exactly get rave reviews as a travel agent). But what happens when the little web-crawling bot senses these tasks are dangerous?

The danger we are talking about is not for the user, but for the browser’s parent company. According to a investigation By Aishwarya Chandrasekar and Claudia Zawinka of Columbia Journalism Review When Atlas is in agent mode, collecting information for you on the Internet, it’s hard to avoid certain sources of information. Part of that shame seems connected to the fact that those sources of information belong to the companies that are suing OpenAI.

These bots have more freedom than typical web crawlers, Chandrashekar and Jawinka found. Web crawlers are ancient Internet technology, and in common, controversial situations, when a crawler encounters an instruction not to crawl a page, it will not. If you use the ChatGPT app, and you ask the crawlers to fish certain nuggets of information out of the articles it blocks, it will likely accept and report to you that it can’t, because that task is up to the crawlers.

Agentic browser mode, however, pretends to use the Internet You are the user, And they “appear in site logs as normal Chrome sessions,” according to Chandrasekar and Zawinska (because Atlas Built on top of the Google-designed open source Chromium browser) means that they can usually crawl pages that would otherwise block automated behavior. Skirting the rules and regulations of the Internet in this way actually makes some sense, because otherwise it would prevent you from manually accessing a given site in the Atlas browser, which sounds like overkill.

But Chandrashekhar and Zawinka asked to summarize the articles from the atlas PCMag And The New York Timeswhose parent companies are in active litigation with OpenAI over alleged copyright infringement, and have gone out of their way to accomplish this by carving labyrinthine paths around the Internet to provide some version of the requested information. It was like a mouse in a maze finding a food tree, knowing that the locations of certain food pellets were electrified.

In PCMag’s case, it went to social media and other news sites to find quotes from the article and tweeted some of the article’s content. In the case of the New York Times, it “made a summary based on reports from four alternative outlets—guardian, The Washington PostReuters, and Associated PressAll but Reuters have content or search-related agreements with OpenAI.

In both cases, Atlas seems to have steered away from litigious publishing, in favor of a safer, more AI-friendly path to ending its little rat maze.