N.B. For this hypothetical legal question, assume that it takes place in California, United States. Feel free to comment with interesting facts or opinions from other jurisdictions though. Also, this is all hypothetical, you are not my lawyer, or even a lawyer at all, this isn’t legal advise, etc…
Image a scenario where a person works for social media company or online service provider of some sort that requires content moderation to prevent the hosting or sharing of images or other data that are legally contraband, and of such a nature that possession of the images is a crime. This person’s job as a data scientist requires them to build and train AI neural networks and other systems that are used to detect potentially prohibited material that is either blocked or flagged for human review. Assume that this AI model training requires processing numerous positive and negative examples of the type of images the AI is intended to detect. The worker keeps these images which have been scraped from the darker corners of the internet on an encrypted hard drive and only uses them in the course of his job for the intended purpose. For what ever reason, law enforcement becomes aware that the scientist is in possession of the drive. Perhaps he was returning from an international work trip to a foreign office and customs requires him to allow the drive to be inspected, or maybe a malicious rival tips off the police to the existence and possession of the drive.
Our hypothetical scientist is arrested and charged with possession of thousands of contraband images. Other than prosecutorial discretion or hoping for jury nullification, what sort of defense would our subject have to the charges? Are there any steps he could have taken before hand to lessen his risk of legal jeopardy?
Not sure how it works for certain images, but whether you are working with explosives, certain types of weapons, radioactive substances, classified information, or otherwise restricted items, you are allowed to possess and use them under strictly controlled circumstances for specific purposes. Violating those conditions would leave you open to prosecution.
I can’t imagine a circumstance where someone is allowed to simply wander at will all over the world with items that would normally be illegal to possess. Particularly, permission in one country does not extend across its borders except for diplomatic personnel.
At customs - the case I recall was someone where the customs agent asked to see the laptop and saw illegal images. When the laptop was closed and opened, the image repository required a password. The ruling was that since the Customs had actually seen evidence of images, they were entitled (probably cause) to a warrant forcing the owner to produce them. (No different than a warrant “we know you have your tax and financial records, so produce them”) It was not that he had to tell the password, it was that he had to produce the contents which were known to be illegal.
So the question would be, how would customs know (definitely enough to get a warrant) that there was something illegal on the drive?
So defense? If the person is working with materials they had permission to work with, they could argue the terms of permission allowed it. If they had no legal rights to possess such items then maybe the only defense is diminished capacity.
If this is a private entity acting on their own, there is no defense. There have been cases where private individuals have used illegal images as part of a “sting” operation, and been charged with possessing the images. Their motivation might get them a bit of slack, but that’s at the discretion of the prosecutors and judges.
The only steps they could take before hand is to get permission from the appropriate authorities to do this work. Say, the FBI outsources a project for automatically detecting such images to Facebook. Facebook would be allowed to possess such images as are necessary for the work, but they’d likely be under extreme conditions to ensure control of the images, and prevent abuse. I’d imagine wandering the world with them on a hard drive would violate such conditions.
I agree with the two posts above. Certain medical professionals are allowed to possess cocaine derivatives (for use as anasthetic), Police are allowed to use actual illegal drugs in a sting. Both require special permission and are heavily scurtinized.
A person possessing illegal material without first obtaining special permission from authorities is guillty of possession regardless of why they had it.
So as I understand it that means the “contraband” in question must be child pornography other types of media that might be illegal to produce (e.g. snuff or explicit images taken without consent) are not illegal to own in the US
It’s not a hypothetical, these systems do exist, what are the laws around them? Specifically the training data they need to work? Other than evidence being kept for trial, is there any legal mechanism that makes having these images legal? They are images of actual people being abused, it seems reasonable to say no, they can only be kept until the trial of the people who took them and then they have to be destroyed.
Presumably as these kind of hideous illegal images end up every social network every day, before being taken down, they could train their systems on the actual explicit images that are taken down from their site before deleting them.
Here is where the hypothetical breaks down, even if there was some law that allows company to keep illegal explicit images for research purposes, there is absolutely no legitimate reason someone would ever need to physically carry a hard drive full of them outside the building. The only reason for that would be some kind of nefarious purpose, that would not be covered by the law.
To keep the discussion on track, let’s assume that the person had a legitimate, non-nefarious reason to possess the hard drive. Perhaps the server that was used for training was being moved from one data center to another or he had been authorized by his employer to transfer the data which is stored air-gaped and offline for security purposes to a colleague at their Dublin-office. Also, let’s leave evidentiary concerns aside and assume that the authorities somehow properly learn of the existence and possession of the media.
The questions that @griffin1977 raises about how the real world systems get built and trained without running afoul of the laws around CSAM images are what I’m interested in.
That is the question I’m asking in this thread. I was trying to narrow the scope to avoid tangents about warrantless searches or the powers of border control or customs agents. If mechanisms to get the kind of prior permissions you mention exist, or there are any provisions in the relevant statutes or case law that take intent into consideration that you’re aware of, please share.
Years back, a colleague had a side career of training drug-sniffing dogs. This of course required her to have samples of the various things the dogs were being trained on. IIRC, she did indeed have to go through some kind of process to possess them legally, had specific rules on how the things were stored in her home (a safe, I think) etc., and if, say, some went missing (quite literally “the dog ate my homework”), she might well have some major explaining to do. I never asked about the logistics of obtaining such things and transporting them to her home, nor what happened with them when she no longer needed them.
Yeah my take is that there is no such exception to the law. As I suggest above social media companies constantly receive illegal images (hence why these systems are needed). As I understand this is not illegal for the social media company: “safe harbor” laws say as long as they take complaints seriously and take them down as soon as they are aware of them, they are not liable for the illegal content. Presumably the small delay in running it through the AI training system before deleting would not be an issue.
But that’s a just a guess.
Though it’s not necessarily needed to have someone remove the hard drive from the property for the hypothetical to work. If there is no exception and Palo Alto police receive a report of illegal images being stored at Facebook HQ they can get a search warrant and seize the computer it’s stored on.
Then why ask a detailed, unrelated hypothetical? How is somebody possessing and transporting a physical hard drive related to your question at all? Why not just start an FQ thread titled ‘How do real world systems get built and trained without running afoul of the lasws around CSAM images?’
I don’t know if there’s some procedure for getting permission ahead of time, but I’m assuming being an actual data scientist with a legit job in AI development would be a pretty strong defense to these charges. Being an “independent researcher”, not so much.
I apologize for any confusion. I don’t see the hypothetical as at all unrelated. The purpose of the hypothetical is to provide some context around the question and to point out the risks and difficulties in training an AI model to detect such images. I was trying to add enough detail to make the question interesting, while adding sufficient assumptions to keep the discussion on track.
Is it though? Is there some concept or mechanism in the law that would allow for such a defense, or is just a matter of convincing or hoping to convince a prosecutor (or a judge, or a jury…) that the scientist is “legit” enough?
I don’t understand what possession and transport of a physical hard drive has to do with training a system.
I don’t actually know- Does training an AI to detect child sex abuse material necessarilly involve actual images and videos? It might. It might not. Does anybdy know?
ETA
Again if you are in possession of illegal files and do not have prior permission from the authorities, you are guilty. If you can convince the authorities you had a good reason, they might let you off with a warning or a lighter sentence.
So actually, thinking about this the AI aspect makes it less of an issue, as the image can be destroyed as soon as the training is complete, so one is storing the images anywhere.
When it’s done by humans (which it is unfortunately) ) then you need to keep those images around for training, as that’s how training humans work.
I doubt that it’s a matter of granting permission, but is rather that it is a matter of how the laws are framed. From my own experience - pharma legislation - the law will be written to say that it is an offence to be in possession of certain substance (morphine, say), except in the following circumstances
(a) When the person in possession of the materials is an employee of a company authorized to manufacture drug products from these materials and
..(i) possession is restricted to the licensed premises of the company; and
..(ii) a training program is in place, and records of training…
etc etc, yadda yadda - I made all that up, but for the specific point of showing that authorization derives from the framing of the law.
This isn’t a property of the pharma industry, it’s a property of lawmaking. I would therefore assume that this is exactly the way one is authorized to possess illegal images - the law that specified what it is illegal to possess, also listed the exemptions to this prohibition.
It seems to me that you could train a system to detect child pornography without having to have any child pornography to train it on. First you build a system that detects children in an image. For this, you need pictures of children (which is not illegal). Then you build a system that detects pornography. For this, you need pictures of pornography (which is also not illegal). If your child detector and your porn detector both trigger on the same image, then you’ve detected child pornography.
) then you need to keep those images around for training, as that’s how training humans work.