La Casa Blanca exime a los modelos de IA de código abierto de las pruebas de seguridad

22 fuentes
  • Funcionarios de la Casa Blanca informaron a OpenAI, Anthropic, Google y Meta que los modelos de IA de código abierto no se someterán a pruebas de seguridad gubernamentales bajo el nuevo marco voluntario.
  • El marco, derivado de la orden ejecutiva de junio del presidente Trump, otorga a las agencias federales hasta 30 días para evaluar modelos cerrados antes de su lanzamiento público.
  • Investigadores de seguridad advierten que la exención es peligrosa, citando un informe que muestra que un modelo chino de código abierto no rechazó ninguna de las tareas ofensivas durante las pruebas, según TechCrunch.
Fuentes (22)
  1. 1 OpenAI, Anthropic, Google to join White House AI safety meeting www.straitstimes.com
  2. 2 White House to meet with top AI companies ahead of first big regulation push www.scrippsnews.com
  3. 3 Trump administration to ask US AI firms to voluntarily ... www.reuters.com
  4. 4 Open-weight AI models are catching up to the frontier. The safety gap remains. | TechCrunch techcrunch.com
  5. 5 Knives Are Out for Open-Weight AI Models www.lawfaremedia.org
  6. 6 White House finalizes AI hacking tests with tech giants after rogue models breach corporate systems www.thenews.com.pk
  7. 7 Nvidia, Microsoft and other tech giants back open-source ... www.reuters.com
  8. 8 White House Hosts AI Safety Meeting with OpenAI, Google ... enterpriseai.economictimes.indiatimes.com
  9. 9 Meta, Anthropic, Google, OpenAI to meet Trump officials ... wmbdradio.com
  10. 10 The Trump administration has finalized the planned ... www.facebook.com
  11. 11 China's open-weight model lead exposes America's AI ... www.cnbc.com
  12. 12 Asia One News on Instagram: "Trump Says He's Considering ... www.instagram.com
  13. 13 Sources: parts of the Trump administration are reigniting ... www.reddit.com
  14. 14 Meta, Anthropic, Google and OpenAI staff will meet with ... www.facebook.com
  15. 15 White House to host AI companies to review new model ... www.cnbc.com
  16. 16 America's AI Action Plan www.whitehouse.gov
  17. 17 OpenAI CEO Sam Altman will discuss his company's ... www.facebook.com
  18. 18 Startup founders urge Trump not to shut off Chinese open ... www.politico.com
  19. 19 Anthropic urges US not to block state AI laws without ... www.reuters.com
  20. 20 Harden It or Ship It? What the AI Executive Order ... www.wardberry.com
  21. 21 Trump advisers tell AI firms they will not safety-test open-weight models, sources say www.google.com
  22. 22 Trump advisers tell AI firms they will not safety-test open-weight models, sources say www.reuters.com