OpenAI previews Astra in Washington as Aug 1 AI deadline nears

OpenAI previewed its new Astra artificial intelligence model to US policymakers, regulators and Trump administration officials in Washington this week, according to The Information, with the government's 60-day AI framework deadline landing Aug. 1.

"The government and private sector have worked together in a way we have never seen before and this foundation of America First is unprecedented," Susie Wiles, White House chief of staff, said in a post on X in late June, as the administration finalized the framework's details.

Altman met with Wiles during his visit and told CNBC he has seen a draft of the framework, adding he did not have "anything specifically in mind" as far as proposed changes. The preview follows a period of tense coordination between OpenAI and the government: OpenAI agreed to limit the rollout of its GPT-5.6 model series to a group of "trusted partners" at the request of US officials, while Anthropic disabled access to its Fable 5 and Mythos 5 models to comply with a Commerce Department export control directive.

The framework, which involves a classified benchmarking process to assess models' cyber capabilities, will determine whether models qualify as "covered frontier models" and which "trusted partners" gain access. The outcome carries direct implications for OpenAI's commercial partners, including Microsoft, and for the broader AI sector as companies navigate the balance between rapid deployment and government oversight.

Open-Weight Debate Splits Silicon Valley

The Astra preview comes as the tech industry's most powerful leaders clash over how to regulate increasingly capable AI models. Nvidia's Jensen Huang released a letter urging policymakers to avoid "premature restrictions" on open-weight models, a position backed by Microsoft, Meta, Palantir and more than 20 other companies. OpenAI joined the letter, but Anthropic declined, instead outlining its position in a separate blog post.

The debate has been sharpened by the release of Moonshot AI's Kimi K3, an open-weight Chinese model that a joint US-UK government assessment concluded is roughly six months behind US models in cyber capabilities. The White House stated that Moonshot trained K3 using banned US AI chips located in Thailand and data obtained through "distillation attacks" on leading US AI labs. Moonshot charges $3 per million tokens to access K3 on its cloud, which is more than certain versions of Anthropic and OpenAI's flagship models.

China Competition Shapes the Regulatory Calculus

The regulatory framework is inseparable from the US-China AI race. Treasury Secretary Scott Bessent characterized Chinese distillation attacks as "IP theft," while CIA Director John Ratcliffe likened AI to "digital nuclear weapons." Chris McGuire, senior fellow at the Council on Foreign Relations, said the United States should cut Chinese models off from US cloud infrastructure and computing power to maintain its technological lead.

For OpenAI, the Astra preview reflects a strategy of proactive regulatory engagement. The company has navigated a murky deployment process since Trump signed the executive order in early June, and a finalized framework could provide clearer rules for bringing frontier models to market. Microsoft, OpenAI's largest commercial partner, stands to benefit from a smoother regulatory path for its AI investments. The framework's classified benchmarking process will also set the standard for how future frontier models from OpenAI, Anthropic and Google are evaluated before public release.

This article is for informational purposes only and does not constitute investment advice.