AI
White House reportedly finalises confidential voluntary pre-release model reviews

it was reported that the administration has finalised a classified framework under which participating advanced-model companies voluntarily submit models for government safety and cybersecurity review up to 30 days before release. Criteria are reportedly shared only with selected firms, and open-source models are excluded. No public framework or primary document was available at cut-off.
Impact: the mechanism could create a de facto assurance layer for frontier releases without legislation. Confidential criteria, voluntary participation and exclusion of open models raise questions about consistency, appeal, competition, transparency and international interoperability.




