Major Sites Are Saying No to Apple’s AI Scraping

This summer, Apple gave websites more control over whether the company could train its AI models on their data. Major publishers and platforms like The New York Times and Facebook have already opted out.

Less than three months after Apple quietly debuted a tool for publishers to opt out of its AI training, a number of prominent news outlets and social platforms have taken the company up on it.

WIRED can confirm that Facebook, Instagram, Craigslist, Tumblr, The New York Times, The Financial Times, The Atlantic, Vox Media, the USA Today network, and WIRED’s parent company, Condé Nast, are among the many organizations opting to exclude their data from Apple’s AI training. The cold reception reflects a significant shift in both the perception and use of the robotic crawlers that have trawled the web for decades. Now that these bots play a key role in collecting AI training data, they’ve become a conflict zone over intellectual property and the future of the web.

This new tool, Applebot-Extended, is an extension to Apple’s web-crawling bot that specifically lets website owners tell Apple not to use their data for AI training. (Apple calls this “controlling data usage” in a blog post explaining how it works.) The original Applebot, announced in 2015, initially crawled the internet to power Apple’s search products like Siri and Spotlight. Recently, though, Applebot’s purpose has expanded: The data it collects can also be used to train the foundational models Apple created for its AI efforts.

Applebot-Extended is a way to respect publishers' rights, says Apple spokesperson Nadine Haija. It doesn’t actually stop the original Applebot from crawling the website—which would then impact how that website’s content appeared in Apple search products—but instead prevents that data from being used to train Apple's large language models and other generative AI projects. It is, in essence, a bot to customize how another bot works.

Publishers can block Applebot-Extended by updating a text file on their websites known as the Robots Exclusion Protocol, or robots.txt. This file has governed how bots go about scraping the web for decades—and like the bots themselves, it is now at the center of a larger fight over how AI gets trained. Many publishers have already updated their robots.txt files to block AI bots from OpenAI, Anthropic, and other major AI players.

Robots.txt allows website owners to block or permit bots on a case-by-case basis. While there’s no legal obligation for bots to adhere to what the text file says, compliance is a long-standing norm. (A norm that is sometimes ignored: Earlier this year, a WIRED investigation revealed that the AI startup Perplexity was ignoring robots.txt and surreptitiously scraping websites.)

Applebot-Extended is so new that relatively few websites block it yet. Ontario, Canada–based AI-detection startup Originality AI analyzed a sampling of 1,000 high-traffic websites last week and found that approximately 7 percent—predominantly news and media outlets—were blocking Applebot-Extended. This week, the AI agent watchdog service Dark Visitors ran its own analysis of another sampling of 1,000 high-traffic websites, finding that approximately 6 percent had the bot blocked. Taken together, these efforts suggest that the vast majority of website owners either don’t object to Apple’s AI training practices are simply unaware of the option to block Applebot-Extended.

In a separate analysis conducted this week, data journalist Ben Welsh found that just over a quarter of the news websites he surveyed (294 of 1,167 primarily English-language, US-based publications) are blocking Applebot-Extended. In comparison, Welsh found that 53 percent of the news websites in his sample block OpenAI’s bot. Google introduced its own AI-specific bot, Google-Extended, last September; it’s blocked by nearly 43 percent of those sites, a sign that Applebot-Extended may still be under the radar. As Welsh tells WIRED, though, the number has been “gradually moving” upward since he started looking.

Welsh has an ongoing project monitoring how news outlets approach major AI agents. “A bit of a divide has emerged among news publishers about whether or not they want to block these bots,” he says. “I don't have the answer to why every news organization made its decision. Obviously, we can read about many of them making licensing deals, where they're being paid in exchange for letting the bots in—maybe that's a factor.”

Last year, The New York Times reported that Apple was attempting to strike AI deals with publishers. Since then, competitors like OpenAI and Perplexity have announced partnerships with a variety of news outlets, social platforms, and other popular websites. “A lot of the largest publishers in the world are clearly taking a strategic approach,” says Originality AI founder Jon Gillham. “I think in some cases, there's a business strategy involved—like, withholding the data until a partnership agreement is in place.”

There is some evidence supporting Gillham’s theory. For example, Condé Nast websites used to block OpenAI’s web crawlers. After the company announced a partnership with OpenAI last week, it unblocked the company’s bots. (Condé Nast declined to comment on the record for this story.) Meanwhile, Buzzfeed spokesperson Juliana Clifton told WIRED that the company, which currently blocks Applebot-Extended, puts every AI web-crawling bot it can identify on its block list unless its owner has entered into a partnership—typically paid—with the company, which also owns the Huffington Post.

Because robots.txt needs to be edited manually, and there are so many new AI agents debuting, it can be difficult to keep an up-to-date block list. “People just don’t know what to block,” says Dark Visitors founder Gavin King. Dark Visitors offers a freemium service that automatically updates a client site’s robots.txt, and King says publishers make up a big portion of his clients because of copyright concerns.

Robots.txt might seem like the arcane territory of webmasters—but given its outsize importance to digital publishers in the AI age, it is now the domain of media executives. WIRED has learned that two CEOs from major media companies directly decide which bots to block.

Some outlets have explicitly noted that they block AI scraping tools because they do not currently have partnerships with their owners. “We’re blocking Applebot-Extended across all of Vox Media’s properties, as we have done with many other AI scraping tools when we don’t have a commercial agreement with the other party,” says Lauren Starke, Vox Media’s senior vice president of communications. “We believe in protecting the value of our published work.”

Others will only describe their reasoning in vague—but blunt!—terms. “The team determined, at this point in time, there was no value in allowing Applebot-Extended access to our content,” says Gannett chief communications officer Lark-Marie Antón.

Meanwhile, The New York Times, which is suing OpenAI over copyright infringement, is critical of the opt-out nature of Applebot-Extended and its ilk. “As the law and The Times' own terms of service make clear, scraping or using our content for commercial purposes is prohibited without our prior written permission,” says NYT director of external communications Charlie Stadtlander, noting that the Times will keep adding unauthorized bots to its block list as it finds them. “Importantly, copyright law still applies whether or not technical blocking measures are in place. Theft of copyrighted material is not something content owners need to opt out of.”

It’s unclear whether Apple is any closer to closing deals with publishers. If or when it does, though, the consequences of any data licensing or sharing arrangements may be visible in robots.txt files even before they are publicly announced.

“I find it fascinating that one of the most consequential technologies of our era is being developed, and the battle for its training data is playing out on this really obscure text file, in public for us all to see,” says Gillham.

Source

RIP Matrix | Farewell my friend  

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of July): 3,313 news posts