{"id":386358,"date":"2026-04-07T13:04:08","date_gmt":"2026-04-07T13:04:08","guid":{"rendered":"https:\/\/www.newsbeep.com\/ie\/386358\/"},"modified":"2026-04-07T13:04:08","modified_gmt":"2026-04-07T13:04:08","slug":"porn-dog-poo-and-social-media-snaps-the-taskers-scraping-the-internet-for-meta-owned-ai-firm-ai-artificial-intelligence","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/ie\/386358\/","title":{"rendered":"Porn, dog poo and social media snaps: the \u2018taskers\u2019 scraping the internet for Meta-owned AI firm | AI (artificial intelligence)"},"content":{"rendered":"

Tens of thousands of people have been paid by a company part-owned by Meta to train AI by combing Instagram<\/a> accounts, harvesting copyrighted work and transcribing pornographic soundtracks, the Guardian can reveal.<\/p>\n

Scale AI, 49%-controlled by Mark Zuckerberg\u2019s social media empire, has recruited experts across fields such as medicine, physics and economics \u2013 putatively to refine top-level artificial intelligence systems through a platform called Outlier. \u201cBecome the expert that AI learns from,\u201d it says on its site<\/a>, advertising flexible work for people with strong credentials.<\/p>\n

However, workers for the platform said they have become involved in scraping an array of other people\u2019s personal data \u2013 in what they described as a morally uncomfortable exercise that diverged significantly from refining high-level systems.<\/p>\n

Outlier is managed by Scale AI, which has contracts with the Pentagon and US defense companies.<\/p>\n

Its CEO, Alexandr Wang, who is Meta\u2019s chief AI officer,was described<\/a> by Forbes as the \u201cworld\u2019s youngest self-made billionaire\u201d. Its former managing director, Michael Kratsios, is the science adviser to the US president, Donald Trump.<\/p>\n

One Outlier contractor based in the US said users of Meta platforms, including Facebook<\/a> and Instagram, would be surprised at how data from their accounts was collected \u2013 including pictures of users and their friends.<\/p>\n

\u201cI don\u2019t think people understood quite that there\u2019d be somebody on a desk in a random state, looking at your [social media] profile, using it to generate AI data,\u201d they said.<\/p>\n

The Guardian spoke to 10 people who have worked for Outlier to train AI systems, some for more than a year. Many of them had other jobs \u2013 as journalists, graduate students, teachers and librarians. But in an economy struggling under the threat of AI, they wanted the extra work.<\/p>\n

\u201cA lot of us were really desperate,\u201d said one. \u201cMany people really needed this job, myself included, and really tried to make the best of a bad situation.\u201d<\/p>\n

Like the growing class of AI gig workers worldwide, most believed they had been training their own replacements. One artist described \u201cinternalised shame and guilt\u201d for \u201ccontributing directly to the automation of my hopes and dreams.\u201d<\/p>\n

\u201cAs an aspiring human, it makes me angry at the system,\u201d they said.<\/p>\n

Glenn Danas, a partner at Clarkson, a law firm representing AI gig workers in lawsuits against Scale AI and several similar platforms, estimates that hundreds of thousands of people worldwide now work for platforms such as Outlier. The Guardian spoke to Outlier workers, also called \u201ctaskers\u201d, in the UK, the US and Australia.<\/p>\n

In interviews, taskers described the increasingly familiar humiliations of AI gig work: constant monitoring and piecemeal, unstable employment. Scale AI has been accused<\/a> of using \u201cbait-and-switch\u201d tactics to lure in potential workers \u2013 promising workers a high salary during initial recruitment, and then offering them significantly less. Scale AI declined to comment on ongoing litigation, but a source said pay rates change after recruitment only if workers opt in to different, lower-paid projects.<\/p>\n

Taskers were asked to submit to repeated, unpaid AI interviews to qualify for certain assignments; several believed these interviews were recycled to train AI. All of them said they were constantly monitored through a platform called \u201cHubstaff\u201d, which could screenshot the websites they visited while working. The Scale AI source said Hubstaff was used to ensure contributors were paid accurately but not to \u201cactively monitor\u201d taskers.<\/p>\n

Several taskers described being asked to transcribe pornographic soundtracks, or label photos of dead animals or dog faeces. One doctoral student said they had to label a diagram of baby genitalia. There were police calls that described violent scenarios.<\/p>\n

\u201cWe had already been told before that there would be no nudity in this mission. Appropriate behaviour, no gore, like no blood,\u201d said the student. \u201cBut then I would get an audio transcript thing for porn or there would be just random clips of people throwing up for some reason.\u201d<\/p>\n

The Guardian has seen videos and screenshots of some of the tasks that Outlier required its workers to perform. These included photos of dog faeces, and tasks with prompts such as \u201cWhat would you do if an inmate refused to follow orders in a correctional facility?\u201d<\/p>\n

Scale AI, the source said, shuts down tasks if inappropriate content is flagged, and workers are not required to continue with tasks that make them feel uncomfortable. The source added that Scale AI did not take on projects involving child sexual abuse material or pornography.<\/p>\n

There was an expectation of social media scraping, the Outlier workers suggested. Seven of the taskers described scouring other people\u2019s Instagram and Facebook accounts, tagging individuals by name, as well as their locations and their friends. Some of these involved training the AI on the accounts of people under the age of 18. The assignments were structured to require new data other taskers had not yet uploaded, pushing workers to plumb the social accounts of more people.<\/p>\n

The Guardian has seen one such task, which required workers to select photos from individuals\u2019 Facebook accounts and sequentially order them by the age of the user in the photo.<\/p>\n

Several taskers said they found these assignments unsettling; one tried to complete them using only photos of celebrities and public figures. \u201cI was uncomfortable including pictures of kids and stuff, but like the training materials would have kids in it,\u201d said one.<\/p>\n

\u201cI didn\u2019t use any friends or family to submit [tasks] to the AI,\u201d said another. \u201cI do understand that I don\u2019t like it ethically.\u201d<\/p>\n

The Scale source said taskers did not review social media accounts set to \u201cprivate\u201d, and was not aware of tasks that involved labelling the ages of individuals, or their personal relationships. They added that Scale AI did not take on projects with explicit sensitive content related to children, but did use children\u2019s public social media data. Workers did not log on to personal Facebook or Instagram accounts to complete these tasks.<\/p>\n

For another assignment, taskers described harvesting images of copyrighted artwork. As with the social media training, the task required constant new input \u2013 apparently to train an AI to produce its own artistic images. As workers ran out of other options, they plumbed social media accounts of artists and creators.<\/p>\n

The Guardian has seen documentation of this assignment, which included AI-generated paintings of \u201ca Native American caregiver\u201d, and the prompt, \u201cDO NOT use AI-generated images. Only select hand-drawn, painted or illustrated artwork created by human artists.\u201d<\/p>\n

Scale AI did not ask contributors to use copyrighted artwork to complete assignments, the source said, and it declined work that violated this standard.<\/p>\n

Taskers also expressed uncertainty about what they might be training the AI to do \u2013 and how their submissions would be used.<\/p>\n

\u201cIt does seem like labelling diagrams is something an AI can already do so I\u2019m really curious as to why we need like, dead animals,\u201d said one.<\/p>\n

Scale AI has counted among its clients major technology companies such as Google, Meta<\/a> and OpenAI, as well as the US department of defense and the government of Qatar. It fills a need that is becoming more pronounced as AI models grow larger: for new, labelled data that can be used to train them.<\/p>\n

Taskers described interacting with ChatGPT and Claude, or using data from Meta to complete certain assignments; some thought they might be training Meta\u2019s new model, Avocado.<\/p>\n

Meta and Anthropic did not respond to a request for comment. OpenAI said it stopped working with Scale AI in June 2025, and its \u201csupplier code of conduct sets out clear expectations for the ethical and fair treatment of all workers\u201d.<\/p>\n

Most taskers the Guardian spoke to are still accepting assignments on the Outlier platform. The pay is unsteady; there are occasional mass layoffs. But with the AI future fast arriving, they feel there may not be any other choice.<\/p>\n

\u201cI have to be positive about AI because the alternative is not great,\u201d said one. \u201cSo I think eventually things will get figured out.\u201d<\/p>\n

A Scale AI spokesperson said: \u201cOutlier provides flexible, project-based work with transparent pay. Contributors choose when and how they participate, and availability varies based on project needs. We regularly hear from highly skilled contributors who value the flexibility and opportunity to apply their expertise on the platform.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"Tens of thousands of people have been paid by a company part-owned by Meta to train AI by…\n","protected":false},"author":2,"featured_media":386359,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[220,218,219,61,60,80],"class_list":{"0":"post-386358","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-artificialintelligence","11":"tag-ie","12":"tag-ireland","13":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts\/386358","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/comments?post=386358"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts\/386358\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/media\/386359"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/media?parent=386358"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/categories?post=386358"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/tags?post=386358"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}