* Porn, dog poo and social media snaps: the ‘taskers’ scraping the internet for Meta-owned AI firm | TheGuardian

[Flor Lynch]

[May be disturbing Content]Message Body
main region
article
figure
Unlabeled 0
figure end
article end
main region end

main region
article
figure
Scale AI logo
figure end
AI (artificial intelligence)

Porn, dog poo and social media snaps: the ‘taskers’ scraping the
internet for Meta-owned AI firm

Scale AI gig workers describe desperation of using people’s personal
profiles and copyrighted work to train AI
article end
main region end

main region
article
Aisha Down
Aisha Down
Notifications off
Tue 7 Apr 2026 14.00 CEST
Listen to this article

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

Scale AI, 49%-controlled by Mark Zuckerberg’s social media empire, has
recruited experts across fields such as medicine, physics and economics
– putatively
to refine top-level artificial intelligence systems through a platform
called Outlier. “Become the expert that AI learns from,” it says on its
site,
advertising flexible work for people with strong credentials.

However, workers for the platform said they have become involved in
scraping an array of other people’s personal data – in what they
described as a morally
uncomfortable exercise that diverged significantly from refining
high-level systems.

Outlier is managed by Scale AI, which has contracts with the Pentagon
and US defense companies.

Its CEO, Alexandr Wang, who is Meta’s chief AI officer,was
described
 by Forbes as the “world’s youngest self-made billionaire”. Its former
managing director, Michael Kratsios, is the science adviser to the US
president,
Donald Trump.

figure
20/meta-ai-agents-instruction-causes-large-sensitive-data-leak-to-employees

figure end
article end
main region end

A circle logo with the message 'Ask Meta AI anything' below it.webp
main region
article
figure
Meta AI agent’s instruction causes large sensitive data leak to employees
Read more
figure end

One Outlier contractor based in the US said users of Meta platforms,
including
Facebook
 and Instagram, would be surprised at how data from their accounts was
collected – including pictures of users and their friends.

“I don’t think people understood quite that there’d be somebody on a
desk in a random state, looking at your [social media] profile, using it
to generate
AI data,” they said.

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 – as
journalists,
graduate students, teachers and librarians. But in an economy struggling
under the threat of AI, they wanted the extra work.

“A lot of us were really desperate,” said one. “Many people really
needed this job, myself included, and really tried to make the best of a
bad situation.”

Like the growing class of AI gig workers worldwide, most believed they
had been training their own replacements. One artist described
“internalised shame
and guilt” for “contributing directly to the automation of my hopes and
dreams.”

“As an aspiring human, it makes me angry at the system,” they said.

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 “taskers”, in the UK,
the US
and Australia.

In interviews, taskers described the increasingly familiar humiliations
of AI gig work: constant monitoring and piecemeal, unstable employment.
Scale AI
has been
accused
 of using “bait-and-switch” tactics to lure in potential workers –
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.

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 “Hubstaff”, 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 “actively monitor” taskers.

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.

“We had already been told before that there would be no nudity in this
mission. Appropriate behaviour, no gore, like no blood,” said the
student. “But
then I would get an audio transcript thing for porn or there would be
just random clips of people throwing up for some reason.”

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 “What would you do if an inmate refused to
follow orders in a correctional facility?”

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.

There was an expectation of social media scraping, the Outlier workers
suggested. Seven of the taskers described scouring other people’s
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.

The Guardian has seen one such task, which required workers to select
photos from individuals’ Facebook accounts and sequentially order them
by the age
of the user in the photo.

Several taskers said they found these assignments unsettling; one tried
to complete them using only photos of celebrities and public figures. “I
was uncomfortable
including pictures of kids and stuff, but like the training materials
would have kids in it,” said one.

“I didn’t use any friends or family to submit [tasks] to the AI,” said
another. “I do understand that I don’t like it ethically.”

The Scale source said taskers did not review social media accounts set
to “private”, 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’s
public social media data. Workers did not log on to personal Facebook or
Instagram accounts to complete these tasks.

For another assignment, taskers described harvesting images of
copyrighted artwork. As with the social media training, the task
required constant new input
– 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.

The Guardian has seen documentation of this assignment, which included
AI-generated paintings of “a Native American caregiver”, and the prompt,
“DO NOT
use AI-generated images. Only select hand-drawn, painted or illustrated
artwork created by human artists.”

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

Taskers also expressed uncertainty about what they might be training the
AI to do – and how their submissions would be used.

“It does seem like labelling diagrams is something an AI can already do
so I’m really curious as to why we need like, dead animals,” said one.

Scale AI has counted among its clients major technology companies such
as Google,
Meta
 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.

Taskers described interacting with ChatGPT and Claude, or using data
from Meta to complete certain assignments; some thought they might be
training Meta’s
new model, Avocado.

Meta and Anthropic did not respond to a request for comment. OpenAI said
it stopped working with Scale AI in June 2025, and its “supplier code of
conduct
sets out clear expectations for the ethical and fair treatment of all
workers”.

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.

“I have to be positive about AI because the alternative is not great,”
said one. “So I think eventually things will get figured out.”

A Scale AI spokesperson said: “Outlier 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.”