White House plans to extend cyber checks to open AI models: new rules for 'open weights'
🇺🇸 The White House plans to extend the voluntary cyber-verification system to open AI models.
🔍 As soon as open models reach the power level of closed systems (like GPT-5.6), they will enter the pre-release evaluation program for 30 days.
⚖️ Officially, this is not licensing, but it will effectively become a security standard.
⚠️ The main goal is to check models before publishing weights, as after release, they escape the developer's control.
📅 Official rules and thresholds will be published in the coming months.
Read more →