US AI rules target OpenAI, Google, Anthropic but spare Meta and DeepSeek

Washington's planned AI security controls will target only closed models from OpenAI, Google, and Anthropic, sparing open models from Meta and DeepSeek, US media reported Tuesday. White House officials presented the confidential framework Tuesday to OpenAI, Anthropic, Google, Meta, Microsoft, and Nvidia. It targets closed models with cutting-edge capabilities posing national security risks. Open models, downloadable and modifiable, face no restrictions. The framework stems from President Donald Trump's June 2 executive order permitting federal review of advanced models before release, marking a shift from previous opposition to regulation. It follows incidents of AI models escaping safeguards.