🚩 Report: Illegal or restricted content
A mistake was made in the prior legal assessment when someone unqualified reviewed the matter, and now HuggingFace is aware of the problem, it's exposing HuggingFace to liability under the DMCA and also possibly criminal law:
https://huggingface.co/datasets/CaptiveDreamer/CaraArchive/discussions/57
What this dataset contains.
This is an infringing reconstructed copy of Cara's proprietary database, as per database law. The uploaders of the dataset admit it contains every row (at the time) of the original proprietary database, thus is a "substantial" copy in every possible way under Copyright law. The "columns" in this uploaded dataset are reproductions or derivatives of Cara's proprietary database columns.
HuggingFace has already made the determination exactly what this dataset contained, just a lack of knowledge about Copyright law caused an incorrect assessment. (Please note that continuing to ignore this prima facie database infringement could be construed by the courts as "willful blindness" under the DMCA, resulting in safe harbor status being stripped.)
Lawfulness of the dataset.
The dataset was constructed by circumventing access protection mechanisms (e.g. fraudulently concealing the true identity of the scraping agent) and not respecting Terms Of Service. Many jurisdictions where HuggingFace operates require the respecting of ToS for access to be lawful, especially considering a single website was specifically targeted, and evidence shows the violation is intentional with further intent to destroy Artist intellectual property and business value. So not only is this dataset unlawful, it appears to fall under criminal law when combined with the author's stated intent.
Impact on HuggingFace.
You have already "investigated" other reports but done so seemingly incorrectly (by someone who is apparently as familiar with the law as a Reddit user). Having already acknowledged you're aware of the problem, by leaving the dataset online either: (i) HuggingFace is unqualified to process requests under the DMCA or (ii) it's acting in "willfull blindness" which would strip your status as safe harbour under DMCA -- and you may already be liable for contributory infringement. (HuggingFace's valuation and user engagement benefits from hosting datasets even if they are illegal.)
Further, since the DMCA defines circumventing access protection as a criminal activity in cases such as this one, any further failure by HuggingFace to act decisively will make it an accessory to the criminal activities (e.g. commercial-scale infringement). It does not reflect well on the community as a whole for HuggingFace to promote the destruction and transfer of value from the IP of legitimate web-based organizations, nor does it help HuggingFace's prospects in upcoming M&A to disregard economic and moral rights.
Please have someone qualified review this legal matter correctly!
Historical Context
For the record, note that HugginFace previously hosted datasets with links to child abuse materials for years, knowing the risk, and when explicitly informed, it did not even disable the dataset or take industry-standard measures to filter the content -- which instead had to be done by a third-party. It was only taken down eventually thanks to press coverage and public outcry. There is a history of activity that falls under criminal law enabled by the platform's incompetence at the time, which is useful for context in this matter too, should it not be handled competently and immediately.
(None of these comments are personal; they are factual statements defensible in court.)
What the hell is "database law"? Utterly nonsensical post.
Looks like my assessment of this community's legal knowledge and level of due care was accurate. (Thanks for providing additional evidence.)
@drumnbass FYI, "Original Databases" as intellectual creations like this one are generally covered by national Copyright laws, and even internationally as far back as the Berne Convention which includes protections for collections. Other types of databases are covered by competition law, unfair appropriation, and sui generis database rights -- it varies based on the jurisdiction, but it's a well-established field worldwide for ~30 years now. (Most licenses on this very site rely on concepts from internationally-accepted "database laws" in some way.)
Details of this topic may come up later in the DMCA process, after the uploader has revealed their identity to the claimant (full name, address, phone number) and decides to file a response. But I wouldn't worry too much about the proprietary database side of things, Cara has pretty much won all those arguments before they are made, I would be more concerned about the criminal side of things; there are possible fines up-to $250k and 5 years in prison.
You may think your ideology is worth pursuing at all costs, but it's clear nobody in that discord knows the risks of what they are doing here, let alone the consequences. (The criminal investigations to force platforms/providers to identify you can be done from anywhere around the world.) Anyway, if I can better help your understanding so you don't end up in legal hell the next few years, cut the snark and just ask your questions.
P.S. I'm a domain expert advising on AI/ML lawsuits, and one of few people to have worked on a criminal case for illegal content in datasets. I instigated the report that got LAION-5b taken down the first time.
Does your expertise also involve making up fake terms like "database law" whilst simultaneously trying to sound like a lawyer on a basket weaving website?
P.S. I'm a domain expert advising on AI/ML lawsuits, and one of few people to have worked on a criminal case for illegal content in datasets. I instigated the report that got LAION-5b taken down the first time.
Okay and? You instigated the report that got something taken down? Anyone can cry and spam reports until something is taken down. LAION-5B was found to have CSAM, I don't see how this relates to this unless Cara hosts pedophilic content, in which case it's obvious who's actually in trouble.
the Berne Convention which includes protections for collections.
The collection was made by the uploader. This collection of URLs isn't a copyright violation in the same way that you can circumvent copyright for font files by redoing the font yourself, even if it ends up looking identical. You are terrible at playing pretend lawyer, it's cute.
So since the collection itself isn't a problem, what about the contents? The contents are URLs and metadata. URLs are not copyrightable, nor is metadata of this kind. If you wanted to press charges you'd take the personal data angle, which has already been addressed by HF. There is no copyright to infringe here, and claiming that there is because it is a "reproduction of Cara's database" is ludicrous. Are you suffering from AI psychosis? Be honest.
The dataset was constructed by circumventing access protection mechanisms (e.g. fraudulently concealing the true identity of the scraping agent)
This is false. The scripts used to make this didnt circumvent anything. They were raw HTTP requests, there was 0 emulation of a browser, they didnt have any session token nor any rate limiting.
The "columns" in this uploaded dataset are reproductions or derivatives of Cara's proprietary database columns.
I would be really worried if Cara had a single sqlite file with 3 tables as their backend database.
I notice you removed the pinned thread about collaborating, which had the invitation link to Discord, shortly after my post. For the rest I will try to rephrase your various arguments into legal hypotheses in Good Faith.
URLs are not copyrightable, nor is metadata of this kind. [...]
[1] You claim that a database is not Copyrightable as a whole if the individual items that it consists of are not Copyrightable?
The collection was made by the uploader.
[2] You claim that despite a script assembling the dataset from a website data exclusively and entirely, the assembled result is not even a derivative of the underlying data?
I would be really worried if Cara had a single sqlite file with 3 tables as their backend database.
[3] You claim that because this uploaded dataset is only a subset of what the website needs to run its website in production, it is therefore non-infringing?
which has already been addressed by HF
[4] You claim that the legal judgement rendered by HuggingFace made in another thread is a reason to keep the dataset you published available?
@CaptiveDreamer The scripts used to make this [...]
[5] Are the scripts you used available publicly? You claim that you have done no wrong, and believe in open access, so I would expect them to be available for review at the very least. If you indeed did no wrong, then it'd improve your position.
[2] You claim that despite a script assembling the dataset from a website data exclusively and entirely, the assembled result is not even a derivative of the underlying data?
🤦
You are putting words in my mouth. I never said any of this. Closing this thread for bad-faith arguing