Updated
Updated · The Washington Post · Jul 13
Trump Administration Weighs U.S. Open-Source AI Limits as Chinese Models Near 6-12 Month Release
Updated
Updated · The Washington Post · Jul 13

Trump Administration Weighs U.S. Open-Source AI Limits as Chinese Models Near 6-12 Month Release

3 articles · Updated · The Washington Post · Jul 13

Summary

  • U.S. officials and AI companies are discussing a capability framework for open-source models that would benchmark American releases against the current strength of leading Chinese open-source systems.
  • The talks are being driven by expectations that Chinese “Mythos-class” models could be freely downloadable within 6 to 12 months, complicating how Washington and the industry set limits for U.S. open models.
  • That framework effort follows broader White House deliberations over an executive order that could delay or restrict advanced open-weight models, reflecting concern over both security risks and damage to the U.S. open-source AI economy.
  • Separately, the report said long-run data-center power demand may be overstated as companies test more efficient data-transfer methods, including specialized conductors and light-wave-based systems.

Insights

Did Anthropic’s push for AI regulation backfire when its own powerful model was restricted?
As tech giants battle for control, can new training methods let open-source AI win the race?
With AI now capable of expert hacking, is a voluntary safety framework enough to prevent disaster?