Un método del MIT detecta modelos de IA entrenados con imágenes de abuso infantil

37 fuentes
  • El MIT, la Universidad de Boston y el grupo de seguridad infantil Thorn desarrollaron una sonda que identifica modelos de IA ajustados para producir material de abuso sexual infantil sin generar ninguna imagen.
  • La técnica introduce ruido aleatorio a través de las capas internas de un modelo, detectando patrones de entrenamiento dañinos; una solución legal, ya que generar CSAM, incluso para pruebas, es ilegal.
  • Los informes de CSAM generados por IA aumentaron a 67.000 en 2024, y los investigadores afirman que el método es lo suficientemente escalable para plataformas que alojan cientos de miles de cargas de modelos al mes.
Fuentes (37)
  1. 1 MIT's New Method Flags AI Models Trained on Child Abuse ... insideai.news
  2. 2 New method aims to keep kids safe from illegal AI- ... techxplore.com
  3. 3 Non-Generative Assessment of Harmful Model ... arxiv.org
  4. 4 [2604.25119] Evaluation without Generation arxiv.org
  5. 5 Detecting AI-Generated CSAM: Why Hash Matching Falls ... casescan.com
  6. 6 publications - Vinith M. Suriyakumar vms-6511.github.io
  7. 7 Enhancing Conspiracy Detection via Instruction-tuned LLMs aclanthology.org
  8. 8 Method teaches generative AI models to locate ... news.mit.edu
  9. 9 Harm without limits: AI child sexual abuse material through ... www.iwf.org.uk
  10. 10 CSAM Classifiers: Find Novel Content with Predictive AI safer.io
  11. 11 Protect your AI model from misuse info.thorn.org
  12. 12 AI-Generated Child Abuse Images Overwhelm Law ... www.bloomberg.com
  13. 13 10: Generative AI – Adapting LLMs with Parameter-Efficient ... www.youtube.com
  14. 14 2026 MIT AI Student Forum genai.mit.edu
  15. 15 Identifying and Eliminating CSAM in Generative ML Training ... purl.stanford.edu
  16. 16 New Report Reveals Unexpected Safety Risks from AI Fine ... cdt.org
  17. 17 Responsible AI | The 2026 AI Index Report - Stanford HAI hai.stanford.edu
  18. 18 AI CSAM: The Growing Digital Threat to Children ourrescue.org
  19. 19 US investigators are using AI to detect child abuse images ... www.technologyreview.com
  20. 20 Vinith M. Suriyakumar scholar.google.com
  21. 21 MIT Gaussian Probing AI Model Audit ... www.facebook.com
  22. 22 Non-Generative Assessment of Harmful Model ... openreview.net
  23. 23 MIT researcher proposes a way to detect nuclear weapons ... www.instagram.com
  24. 24 ‪Vinith M. Suriyakumar‬ - ‪Google Scholar‬ scholar.google.co.th
  25. 25 Europe revives law allowing big tech to scan for CSAM therecord.media
  26. 26 MIT Researchers Develop a New Way to Detect ... newtoeducation.com
  27. 27 Thorn's technical innovation builds a safer internet www.thorn.org
  28. 28 Advancing Child Safety Online – THORN safeonline.global
  29. 29 https://scholar.google.com/citations?view_op=view_... scholar.google.com
  30. 30 Study proposes framework for 'child-safe AI' following ... techxplore.com
  31. 31 Report calls for AI toy safety standards to protect young ... techxplore.com
  32. 32 How AI is reshaping society and the risks to our kids. www.linkedin.com
  33. 33 and they have the backing of one of the world's biggest AI ... www.facebook.com
  34. 34 Tech Xplore news sciencex.com
  35. 35 Tech Xplore - Technology and Engineering news techxplore.com
  36. 36 Tech Xplore (@TechxploreCom) www.facebook.com
  37. 37 How can we keep children safe as AI reshapes the internet? www.threads.com