White House to vet closed-source AI models before release: Report

The White House has decided that closed-source advanced AI models must be tested by the government for at least 30 days before public release, according to Axios. Open-weight models such as Meta's Llama, Google's Gemma, and Microsoft's Phi will likely be exempt. The move follows a meeting between President Donald Trump and AI companies including OpenAI, Anthropic, and Google, and aims to curb national security risks.

The White House has reportedly decided that closed-source advanced AI models should be tested by the government before being rolled out to the public. While a formal statement from the White House remains awaited on the decision, reports said open-weight AI models like Meta’s Llama Series, Google’s Gemma Series, and Microsoft’s Phi Series will most likely be exempted from the vetting phase. U.S. President Donald Trump met with AI players including OpenAI, Anthropic, and Google at the White House this week. The agenda of this meeting was to review this AI-testing framework which was pitched as a solution to curb frontier AI models from reaching the wrong hands that could threaten national, social, or the financial security of the U.S. AI companies may soon have to grant the White House an early access to their closed-source AI models for internal government trials at least 30 days to test a new model for safety risks before it goes public. However, the government will not be allowed to bar AI companies from pay for a license, get the White House permission, or wait for the official approval before launching their products, Axios reported on Tuesday. The White House does not intend on laying out its AI testing blueprint on public domains over security concerns. Axios did, however, report that all models that enter White House’s testing period would be vetted in high-security environments and up-to-date access logs. A group of government officials will reportedly be part of the AI review process instead of having a single agency complete the testing. As of now, it remains unclear as to what parts of AI models could be considered as a national security risk. White House’s reported decision of leaving open weight AI models out of this testing regime has come just days after industry giants like Nvidia, Microsoft, Meta, IBM, Perplexity, Hugging Face, Dell Technologies, and Y Combinator called for easier rules on these models. The companies collectively pointed out https://thecoinheadlines.com/tech-and-ai/huang-zuck-musk-nadella-dorsey-bat-for-open-source-ai-but-why/article-27254/ that open-weight AI models, that are publicly accessible for anybody to view, use, study, modify, and distribute their codes, are key building blocks for the future of AI. AI firms have been directed by the White House to share models with the government that are as close to public release as possible, Axios said citing sources familiar with the matter. The U.S. is gradually shifting its focus to tighten the noose around the ongoing AI developments. After Anthropic’s Claude Mythos proved to be stronger than existing cybersecurity provisions and OpenAI’s experimental AI model https://thecoinheadlines.com/tech-and-ai/chatgpt-under-outage-after-openais-ai-going-rogue-experience-report/article-26738/ attempting to break out of a sandbox owing to self-training — the White House has seemingly rushed to finalize these internal vetting rules for AI giants to adhere to. The U.S. is looking to keep its AI dominance against competition intensifying from other parts of the world including China and the EU. This is why President Trump’s administration is taking pro-active steps to increase its say in U.S.’ AI development. OpenAI is reportedly mulling if it should give a five percent company stake https://thecoinheadlines.com/tech-and-ai/openai-mulls-giving-five-percent-stake-to-white-house-eyes-public-wealth-fund/article-24496/ to the Trump administration for an estimate of over $42 billion. In exchange, the AI company has pitched for a public wealth fund to distribute profits generated through AI towards public welfare.