Trump's new AI testing framework: Open models spared

The White House presented its new testing framework for powerful AI models to representatives of leading AI companies for the first time on Tuesday. According to US media reports, open-weight models are to remain exempt from security checks. The decision follows weeks of debate within the Trump administration over how to handle open AI models. According to Reuters, representatives from OpenAI, Anthropic, Google, Meta, and Nvidia participated in the discussions. As Wired reports, developers can submit their models for testing up to 30 days before their planned release. Government agencies will then evaluate their cyber capabilities using a secret benchmark system and share the models with federal agencies and selected corporate partners. Participation is voluntary, but opting out could carry significant risks for companies in the event of later security incidents. According to Axios, the framework defines a covered frontier model as a closed system with cutting-edge capabilities and risks to national security, but specific thresholds are missing. This would primarily affect models from OpenAI and Anthropic, while less powerful systems and open-weight models would initially be excluded. The White House is keeping the testing framework under wraps for now. According to media reports, the details are to be accessible only to the participating companies.