No abuse of a real minor is needed.
I suspect none of us are lawyers with enough legal knowledge of the French law to know the specifics of this case
Strange that there was no disagreement before "AI", right? Yet now we have a clutch of new "definitions" all of which dilute and weaken the meaning.
> In Sweden for instance, CSAM is any image of an underage subject (real or realistic digital) designed to evoke a sexual response.
No corroboration found on web. Quite the contrary, in fact:
"Sweden does not have a legislative definition of child sexual abuse material (CSAM)"
https://rm.coe.int/factsheet-sweden-the-protection-of-childr...
> If you take a picture of a 14 year old girl (age of consent is 15) and use Grok to give her bikini, or make her topless, then you are most definately producing and possessing CSAM.
> No abuse of a real minor is needed.
Even the Google "AI" knows better than that. CSAM "is considered a record of a crime, emphasizing that its existence represents the abuse of a child."
Putting a bikini on a photo of a child may be distasteful abuse of a photo, but it is not abuse of a child - in any current law.
Are you from Sweden? Why do you think the definition was clear across the world and not changed "before AI"? Or is it some USDefaultism where Americans assume their definition was universal?
No. I used this interweb thing to fetch that document from Sweden, saving me a 1000-mile walk.
> Why do you think the definition was clear across the world and not changed "before AI"?
I didn't say it was clear. I said there was no disagreement.
And I said that because I saw only agreement. CSAM == child sexual abuse material == a record of child sexual abuse.
So you cant speak Swedish, yet you think you grasped the Swedish law definition?
" I didn't say it was clear. I said there was no disagreement. "
Sorry, there are lots of different judical definitions about CSAM in different countries, each with different edge cases and how to handle them. I very doubt it, there is a disaggrement.
But my guess about your post is, that an American has to learn again there is a world outside of the US with different rules and different languages.
I guess you didn't read the doc. It is in English.
I too doubt there's material disagreement between judicial definitions. The dubious definitions I'm referring to are the non-judicial fabrications behind accusations such as the root of this subthread.
Please don't use the "knowledge" of LLMs as evidence or support for anything. Generative models generate things that have some likelihood of being consistent with their input material, they don't "know" things.
Just last night, I did a Google search related to the cell tower recently constructed next to our local fire house. Above the search results, Gemini stated that the new tower is physically located on the Facebook page of the fire department.
Does this support the idea that "some physical cell towers are located on Facebook pages"? It does not. At best, it supports that the likelihood that the generated text is completely consistent with the model's input is less than 100% and/or that input to the model was factually incorrect.
It has been since at least 2012 here in Sweden. That case went to our highest court and they decided a manga drawing was CSAM (maybe you are hung up on this term though, it is obviously not the same in Swedish).
The holder was not convicted but that is besides the point about the material.
This one?
"Swedish Supreme Court Exonerates Manga Translator Of Porn Charges"
https://bleedingcool.com/comics/swedish-supreme-court-exoner...
It has zero bearing on the "Putting a bikini on a photo of a child ... is not abuse of a child" you're challenging.
> and they decided a manga drawing was CSAM
No they did not. They decided "may be considered pornographic". A far lesser offence than CSAM.
https://www.regeringen.se/contentassets/5f881006d4d346b199ca...
> Även en bild där ett barn t.ex. genom speciella kameraarrangemang framställs på ett sätt som är ägnat att vädja till sexualdriften, utan att det avbildade barnet kan sägas ha deltagit i ett sexuellt beteende vid avbildningen, kan omfattas av bestämmelsen.
Which translated means that the children does not have to be apart of sexual acts and indeed undressing a child using AI could be CSAM.
I say "could" because all laws are open to interpretation in Sweden and it depends on the specific image. But it's safe to say that many images produces by Grok are CSAM by Swedish standards.
You should realize that children have committed suicide before because AI deepfakes of themselves have been spread around schools. Just because these images are "fake" doesn't mean they're not abuse, and that there aren't real victims.
You would be _amazed_ at the things that people commit to email and similar.
Here's a Facebook one (leaked, not extracted by authorities): https://www.reuters.com/investigates/special-report/meta-ai-...
What may they find, hypothetically? Who knows, but maybe an internal email saying, for instance, 'Management says keep the nude photo functionality, just hide it behind a feature flag', or maybe 'Great idea to keep a backup of the images, but must cover our tracks', or perhaps 'Elon says no action on Grok nude images, we are officially unaware anything is happening.'
I mean, perhaps it's time to completely drop these US-owned, closed-source, algo-driven controversial platforms, and start treating the communication with the public that funds your existence in different terms. The goal should be to reach as many people, of course, but also to ensure that the method and medium of communication is in the interest of the public at large.
I think we are getting very close the the EU's own great firewall.
There is currently a sort of identity crisis in the regulation. Big tech companies are breaking the laws left and right. So which is it?
- fine harvesting mechanism? Keep as-is.
- true user protection? Blacklist.
Who decides what communication is in the interest of the public at large? The Trump administration?
I suppose the answer, if we're serious about it, is somewhat more nuanced.
To begin, public administrations should not get to unilaterally define "the public interest" in their communication, nor should private platforms for that matter. Assuming we're still talking about a democracy, the decision-making should be democratically via a combination of law + rights + accountable institutions + public scrutiny, with implementation constraints that maximise reach, accessibility, auditability, and independence from private gatekeepers. The last bit is rather relevant, because the private sector's interests and the citizen's interests are nearly always at odds in any modern society, hence the state's roles as rule-setter (via democratic processes) and arbiter. Happy to get into further detail regarding the actual processes involved, if you're genuinely interested.
That aside - there are two separate problems that often get conflated when we talk about these platforms:
- one is reach: people are on Twitter, LinkedIn, Instagram, so publishing there increases distribution; public institutions should be interested in reaching as many citizens as possible with their comms;
- the other one is dependency: if those become the primary or exclusive channels, the state's relationship with citizens becomes contingent on private moderation, ranking algorithms, account lockouts, paywalls, data extraction, and opaque rule changes. That is entirely and dangerously misaligned with democratic accountability.
A potential middle position could be ti use commercial social platforms as secondary distribution instead of the authoritative channel, which in reality is often the case. However, due to the way societies work and how individuals operate within them, the public won't actually come across the information until it's distributed on the most popular platforms. Which is why some argue that they should be treated as public utilities since dominant communications infrastructure has quasi-public function (rest assured, I won't open that can of worms right now).
Politics is messy in practice, as all balancing acts are - a normal price to pay for any democratic society, I'd say. Mix that with technology, social psychology and philosophies of liberty, rights, and wellbeing, and you have a proper head-scratcher on your hands. We've already done a lot to balance these, for sure, but we're not there yet and it's a dynamic, developing field that presents new challenges.
I don't love heavy-handed enforcement on speech issues, but I do really like a heterogenous cultural situation, so I think it's interesting and probably to the overall good to have a country pushing on these matters very hard, just as a matter of keeping a diverse set of global standards, something that adds cultural resilience for humanity.
linkedin is not a replacement for twitter, though. I'm curious if they'll come back post-settlement.
Libel must be as assertion that is not true. Photoshopping or AIing someone isn't an assertion of something untrue. It's more the equivalent of saying "What if this is true?" which is perfectly legal
Durov was held on suspicion Telegram was willingly failing to moderate its platform and allowed drug trafficking and other illegal activities to take place.
X has allegedly illegally sent data to the US in violation of GDPR and contributed to child porn distribution.
Note that both are directly related to direct violation of data safety law or association with a separate criminal activities, neither is about speech.
There's someone who was being held responsible for what was in encrypted chats.
Then there's someone who published depictions of sexual abuse and minors.
Worlds apart.
If you're going to make serious accusations like that you're going to need to provide some evidence.
“Probably just Talulah and me. What day/night will be the wildest party on your island?” Musk replied, in an apparent reference to his former wife Talulah Riley.
https://www.theguardian.com/technology/2026/jan/30/elon-musk...
I think there's just as much evidence Clinton did as Musk. Gates on the other hand.
Has the latest release changed that narrative?
Elon didn't ask to go, he was invited multiple times
Censorship increases homogeneity, because it reduces the amount of ideas and opinions that are allowed to be expressed. The only resilience that comes from restricting people's speech is resilience of the people in power.
Why isn't that a major red flag exactly?
(it’ll be interesting to see if this discussion is allowed on HN. Almost every other discussion on this topic has been flagged…)
As mentioned in the article, the UK's ICO and the EC are also investigating.
France is notably keen on raids for this sort of thing, and a lot of things that would be basically a desk investigation in other countries result in a raid in France.
When notified, he immediately:
* "implemented technological measures to prevent the Grok account from allowing the editing of images of real people in revealing clothing" - https://www.bbc.co.uk/news/articles/ce8gz8g2qnlo
* locked image generation down to paid accounts only (i.e. those individuals that can be identified via their payment details).
https://www.bbc.com/news/articles/c98p1r4e6m8o
> Have the other AI companies followed suit? They were also allowing users to undress real people
No they weren’t? There were numerous examples of people feeding the same prompts to different AIs and having their requests refused. Not to mention, X was also publicly distributing that material, something other AI companies were not doing. Which is an entirely different legal liability.
“Sorry I broke the law. Oops for reals tho.”
The onus is on the contractor to make sure any classified information is kept securely. If by raiding an office in France a bunch of US military secrets are found, it would suggest the company is not fit to have those kind of contracts.
https://www.the-independent.com/news/world/americas/crime/us...
This step could come before a police raid.
This looks like plain political pressure. No lives were saved, and no crime was prevented by harassing local workers.
The company made and released a tool with seemingly no guard-rails, which was used en masse to generate deepfakes and child pornography.