Technologie · CSAM reports hit 440,000 · ArchivesGuterres dévoile un engagement en trois règles pour la sécurité des enfants face à l'IA — et une question ouverte sur ses signatairesRédigé 7 septembre 2026 à 21:22 UTCRegarderTranscription0:00Secretary General Guterres did not stop0:02at the battlefield. [music]0:03In the same speech in Geneva, he turned0:06to a threat that is already inside0:08millions of homes, and it is targeting0:10children.0:12Guterres unveiled an AI child safety0:15pledge [music] today. Three rules he0:17says every AI company must follow before0:19a single child can access [music] its0:21systems. Three rules. Rule one, prove it0:25is safe before a child ever touches it.0:27Rule two, zero tolerance for0:29AI-generated sexual abuse images. Rule0:32three,0:33>> [music]0:33>> when a child is in distress, the machine0:35must stop and connect them to a real0:37person. His framing was blunt. Guterres0:40said we do not let medicine reach a0:42child [music]0:43until it is proven safe. We test every0:45toy. Yet AI has reached our children0:48before anyone asked what it would0:49[music] do to them. The line that is0:51likely to outlast the speech itself.0:56>> [music]0:57>> Here is why the urgency. The numbers1:00behind AI-generated child sexual abuse1:03material are not rising. They are1:06exploding. In 2023, [music]1:09the National Center for Missing and1:11Exploited Children received [music]1:124,7001:14reports.1:15In 2024,1:17that number [music] hit 67,000.1:20In just the first half of 2025, it1:23reached [music] 440,000,1:25a more than 9,000% increase in 2 years.1:29The Internet Watch Foundation reported a1:31260-fold increase in AI-generated abuse1:35videos this year compared to last.1:37>> [music]1:37>> And the technology is not just producing1:39more abuse content, it is producing1:41worse content.1:42>> [music]1:43>> 65% of AI-generated abuse material is1:46classified as category A, the most1:48extreme [music] depicting rape and1:50sexual torture. For non-AI material,1:53that figure is 43%. [music]1:55The victims are overwhelmingly girls,1:5797%.1:59One in 17 teens worldwide2:01>> [music]2:01>> has already been a direct victim of2:03deepfake nudes. Roughly 300 million2:07children globally are affected by2:09technology-facilitated2:10sexual exploitation [music] and abuse2:12every year. The pledge builds on a joint2:15statement adopted in January, co-signed2:18by 14 UN entities, including UNICEF, and2:21[music] backed by more than 60 civil2:23society organizations. Sophie Kiladze,2:26chair of the Child Rights Committee,2:28called it a historic unified [music]2:30position.2:31Guterres made the announcement here at2:34the UN Global Dialogue on AI governance2:37in Geneva on July 7th.2:40The pledge is not a binding treaty.2:42[music]2:43It is a call to action and a political2:45benchmark.2:46Guterres is betting that naming the2:48standard publicly [music]2:49and attaching the moral weight of his2:51office to it will make it harder for2:53companies and governments to claim they2:54did not know what was happening.2:56The question now is whether any major AI2:58company will sign on [music] and what3:00consequences follow for those that do3:02not.3:03We will be tracking it.Lire l’articleChronologie de l’histoireHeyDex
