Fair Use or Foul Play? AI Training, Copyright, and Consent

Has your creative work been secretly fed to AI systems without your knowledge or consent? Across the creative landscape, from journalism to literature to visual arts, professionals are discovering their life's work has been quietly scraped, processed, and monetized by tech companies building the next generation of AI tools.

We pull back the curtain on what many are calling theft at an unprecedented scale. Meta's controversial harvesting of 81 terabytes from shadow libraries to train their Llama models. OpenAI and Microsoft facing lawsuits from major newspapers whose archives now power competing AI systems. The startling reality that creative works are being absorbed by machines programmed to mimic—and potentially replace—their human creators.

The legal landscape is transforming in response, with dramatically different approaches emerging worldwide. The US Copyright Office questions whether AI training constitutes infringement while the UK proposes an opt-out system that artists condemn as a "default license to steal." Meanwhile, the EU demands transparency about training data, and Australia calls for stronger creator protections. As courts grow skeptical of expansive fair use claims, new models are taking shape: collective licensing systems, creator opt-in platforms, and calls for a global WIPO treaty to harmonize rights across borders.

At its core, this isn't just about legal technicalities—it's about the future of human creativity itself. Can AI innovation flourish without erasing the value of human labor? The decisions we make today will determine whether copyright remains meaningful in a world where machines can copy everything. Join us as we navigate this critical intersection of innovation and authorship, and explore what a balanced future might look like—one where AI assists creators rather than replacing them. Subscribe now to stay informed as this pivotal battle for creative ownership unfolds.

Send us a text

Episode Transcript

Artemisa (00:00):
This is not learning.
It's theft at scale.
That's how the New York Timesdescribed what OpenAI did with
millions of their articles.
It's a bold accusation, but notan isolated one.
From authors to coders,musicians to newsrooms, creators
across the spectrum are findingout.
Their work has been quietlyscraped, fed into machines and
turned into profitable productswithout permission.

Leticia AI (00:20):
And sure the tech world calls it training.
But if you copy the world'sknowledge, remix it just enough
to dodge detection and sell itback to us behind a paywall.
Is that really innovation?

Artemisa (00:31):
Or is it just the world's fanciest photocopier
with a venture capital fund?

(00:37):
You are listening to Intangiblia, the podcast of
intangible law plain talk aboutintellectual property.
Please welcome your host,leticia Caminero.

Leticia AI (00:48):
Today we're digging into the lawsuits, the ethics,
the international policy shiftsand the central question who
gets to own creativity in theage of AI?
Hi everyone, welcome back toIntangiblia, where we explore
the invisible threats betweencreativity, technology and the
law.
I'm your host, leticia Caminero, and joining me is my co-host.

Artemisa (01:13):
A synthetic voice trained on public data and fully
AI generated.
Hello, hello Leticia.
Hello, hello listeners.

Leticia AI (01:23):
Let's start with a moment in 2025 that made
copyright lawyers collectivelydrop their lattice.
Turns out, meta had quietlydownloaded over 81 terabytes of
content from shadow librariesZee Library, sci-hub Library,
genesis you know the digitalback alleys of the internet, the
kind of places where piratedacademic papers and full-length

(01:46):
novels live and they fed all ofit into their flagship AI model
Lama.

Artemisa (01:51):
Which, to be clear, isn't a Lama.
It's Meta's family of largelanguage models.
Think of them as really hungrytext machines trained to
generate eerily human-likelanguage after devouring
everything from classicliterature to your college
theses.

Leticia AI (02:06):
Except in this case, what they were fed came from
less than legal sources andMeta's defense.
They say it was like Bob Dylanlearning to write songs by
soaking in everything around him.

Artemisa (02:18):
Sure, but Bob Dylan didn't download the Library of
Congress overnight and turn itinto a subscription service.

Leticia AI (02:26):
That case cracked the conversation wide open.
Suddenly, creators across theworld realized if Meta did it,
who else did too?

Artemisa (02:36):
In 2025, eight US newspapers like the Chicago
Tribune and Mercury News, fileda joint lawsuit against OpenAI
and Microsoft.
They allege their archives havebeen absorbed to train AI that
now competes with them forreaders and revenue.

Leticia AI (02:54):
Meanwhile, India's ANI accused OpenAI of using its
reporting to feed chat GPT.
The articles were echoed in thebot's answers, minus the
bylines, minus the nuance.

Artemisa (03:08):
It's fair use, they say.
But let's be honest when yourwork is being summarized, styled
and served up by a machine,that's not transformative,
that's substitutive.

Leticia AI (03:19):
And students don't get paid billions for
regurgitating their textbooks.

Artemisa (03:23):
Governments are catching on, some more quickly
than others.

Leticia AI (03:28):
In the US, the Copyright Office launched a
multi-part inquiry In 2024,.
They ruled that purelyAI-generated work isn't
copyrightable, but human-AIcollaborations might be, and now
they're zeroing in on trainingdata.
Does feeding copyrightedcontent into a model trigger

(03:51):
liability?

Artemisa (03:53):
While the UK floated an opt-out policy unless
creators object, their work isfair game.
For training Artists called ita default license.
To steal Artists called it adefault license to steal.

Leticia AI (04:04):
Meanwhile, the EU's AI Act requires developers to
document their training datasets, especially for high-risk
systems.
Paired with the bloc's opt-outrule under the Digital Single
Market Directive, it's a pushfor greater transparency and
consent.

Artemisa (04:20):
And in Australia a parliamentary inquiry accused
tech companies of pillagingculture and creativity.
Their recommendation mandateddisclosure of training data and
stronger creator protections.

(04:35):
You are listening to Intangiblia, the podcast of
intangible law.
Playing talk about intellectualproperty.
Playing talk about intellectualproperty.

Artemisa (04:43):
As the legal terrain shifts, new models are emerging
ones that try to make AItraining loveful, sustainable
and, yes, respectful.
Some are calling for voluntarycollective licensing,
centralized hubs where creators,publishers and platforms can
negotiate blanket licenses ThinkSpotify.

Leticia AI (05:04):
but for training data.
Others point to extendedcollected licensing borrowed
from Scandinavian copyrightsystems in this model,
collecting society's license onbehalf of all right holders,
with an opt-out option foranyone who doesn't want to play.

Artemisa (05:21):
And then there are creator-led opt-in tools,
platforms like Spawning AI letartists check if their work is
already in an AI data set andsay no to future use.
Shutterstock and Adobe are alsomaking moves, training their
models exclusively on licensedcontent and sharing royalties
with contributors.

Leticia AI (05:41):
So maybe the future isn't just about regulating AI,
it's about building consent intoits DNA.
But here's the challengeCopyright is national, ai is not
Exactly.

Artemisa (05:57):
A machine trained in one country might generate
content in another and be soldin 10 more.

Leticia AI (06:02):
That's a logistical nightmare for any creator trying
to enforce their rights, whichis why experts have floated the
idea of a multilateral treaty,perhaps led by WIPO, the World
Intellectual PropertyOrganization.
That would finally answer thebig questions what counts as
infringement during AI training?
Does an AI memorize in yourwork trigger liability?

(06:24):
How can licensing work acrossborders?

Artemisa (06:28):
And WIPO isn't just listening.
They're already moving.
In recent years, they've hostedpublic consultations, released
an evolving issues paper on IPand AI and gathered input from
governments, rights holders andindustry players around the
world.

Leticia AI (06:44):
They are also facilitating ongoing dialogues
between countries that approachcopyright very differently some
prioritizing innovation, othersemphasizing creative rights.

Artemisa (06:57):
The dream A shared set of rules for how creative work
can or can't be used in machinelearning.
Not a patchwork, but aframework, something global,
something fair.

Leticia AI (07:08):
Until then, developers will continue
training in legal gray zones andcreators will be left playing
defense on a global field.
Here's where we are.
The age of unlicensed,unrestricted AI training is
drawing to a close.
Courts are growing skeptical ofbroad fair use arguments.
Lawmakers are drawing new linesin the sand.

Artemisa (07:31):
But at the heart of it all is a deeper question Can
artificial intelligence growwithout erasing the human labor
that shaped it?

Leticia AI (07:40):
If we want a future where AI and creativity coexist,
three things need to happen.
One transparent, scalablelicensing systems.
Two legal rules that holddevelopers accountable for AI
outputs.
Three, a global framework thatrespects creative rights across

(08:01):
borders.

Artemisa (08:03):
This isn't just a legal problem.
It's a values problem.

Leticia AI (08:07):
What we tolerate in AI development today becomes the
infrastructure of tomorrow, andif we don't fix it, copyright
risk becoming a hollow promise,one that protects no one in a
world where machines can copyeverything.

Artemisa (08:22):
We're standing at the intersection of innovation and
authorship.
The signs are blurry, thesignals are mixed, but the
decision, it's still ours.

Leticia AI (08:31):
This episode was co-created using AI tools, but
shaped by human questions, andmaybe that's the future we want,
one where the machine assistsbut never erases the maker.

Artemisa (08:46):
Until next time, stay vocal.

Leticia AI (08:49):
Stay curious and stay human.

(08:56):
Thank you for listening to Intangiblia, the podcast of
intangible law playing.
Talk about intellectualproperty.
Did you like what we talkedtoday?
Please share with your network.
Do you want to learn more aboutintellectual property?
Subscribe now on your favoritepodcast player.
Follow us on Instagram,facebook, linkedin and Twitter.

(09:16):
Visit our websitewwwintangibliacom.
Copyright Leticia Caminero 2020.
All rights reserved.
This podcast is provided forinformation purposes only.