Symbolic imageWhite House AI security review to exempt open-weight models
The White House is readying a voluntary security review for AI models that would leave open-weight systems outside mandatory government checks. US outlets reported the exemption on Wednesday.
What happened
- The New York Times reports the review is voluntary and covers closed-source models, excluding those that publish the underlying code.
- Bloomberg reports that Chinese open-weight models will likewise fall outside US safety testing.
- The White House met OpenAI, Anthropic, Google, Meta and other developers on Tuesday, 4 August, on security testing.
- Alibaba released its most capable model to date, Qwen3.8-Max, with downloadable weights on Monday, 3 August.
- A research firm puts DeepSeek's V4-Flash at more than 100 times cheaper than Anthropic's Claude Fable 5.
The view from outside
All four reports come from US outlets; from Hong Kong, the South China Morning Post covered the Tuesday meeting, and no Chinese official response to the planned exemption appears in the available reporting.
What could happen next
- The framework text is unpublished: scope, timetable and which agency would run the reviews are not reported.
- Whether leading US developers join the voluntary process, and what follows if they decline, is open.
- The New York Times describes the administration's line on open models as shifting, leaving the rules unsettled.