2026-07-02 14:50:57
AI firms brace for pauses, reviews, halt orders as federal AI regulation takes shape
最阅读新闻
The Trump administration is finalizing voluntary standards for frontier AI model releases after months of executive orders, export controls, and forced model shutdowns. Ordinarily, it should come as bullish news as AI firms are getting what they wanted after lobbying Washington since 2025 to preempt a growing patchwork of state AI laws. The federal government swung into action late last year with an executive order that established an AI Litigation Task Force to challenge state regulations and called on Congress to create a single federal standard. The government then put a set of requirements, such as classified benchmarking requirements and pre-release government review periods. AI companies saw this machinery in action after Anthropic was ordered to pull two flagship models offline within days of launch. What is the current federal framework for AI? The Trump administration’s AI policy is carried by three executive orders, with the first, which was signed in December 2025, directing the Commerce Department to evaluate state AI laws and identify those conflicting with federal policy. It also created the litigation task force to sue states whose regulations the administration deemed too burdensome. The second order , which came in June 2026, tasked the National Institute of Standards and Technology (NIST) with building a classified benchmarking process for frontier AI models within 60 days. It also introduced a pre-release evaluation window of up to 30 days of government access before covered frontier models can be shared with trusted partners. Also, the administration stated that no new federal AI regulator will be created as oversight runs through existing agencies, which are mainly the NIST and the Cybersecurity and Infrastructure Security Agency. Now the US government is reportedly in advanced talks with AI companies to finalize voluntary standards for model releases, with an announcement said to be coming soon. The standards would set benchmarks for advanced models and clarify who can access them domestically and abroad. The US government has had hands in directing AI product launches On June 12, Commerce Secretary Howard Lutnick ordered Anthropic to block all foreign access to its Claude Fable 5 and Claude Mythos 5 models just three days after their launch. Anthropic initially announced that it was pausing access to those models to non-US citizens indefinitely, but later stated that it could not enforce nationality-based restrictions and shut both models down for everyone. The company also pushed back on the security concerns claims, calling the jailbreak “narrow” and stating that OpenAI’s GPT-5.5 could find similar software flaws without any bypass. Anthropic said it had conducted thousands of hours of red-teaming with the U.S. government, the UK AI Safety Institute, and private groups before launch. Anthropic has since regained limited permission to sell Mythos 5 to more than 100 approved U.S. organizations, and Fable 5 access has been restored. OpenAI has faced similar constraints, as it had to delay a full public launch of GPT-5.6 at the government’s request, limiting access to roughly 20 companies with individual government sign-off. Google has also been in discussions with officials ahead of releasing advanced coding models with stronger cyber capabilities than previous generations. The state-law problem has not gone away The federal preemption push was supposed to simplify compliance for AI companies. The December 2025 executive order specifically targeted Colorado’s algorithmic discrimination law and called state-level regulation a barrier to innovation. The order directed the Commerce Department to publish an evaluation of state AI laws within 90 days and flag those requiring models to “alter their truthful outputs.” But preemption requires congressional action, and that legislation has not materialized. The states have not given up, as some have continued to forge ahead with their own AI regulations. Companies that have been actively gunning for rules that are all-encompassing in the US are now faced with the state-level rules they wanted to escape and a federal apparatus that has proven willing to suspend products, gate releases, and impose export controls on short notice. Both Anthropic and OpenAI are preparing for IPOs. Anthropic filed a confidential S-1 with the SEC on June 1 after raising $65 billion at an implied $965 billion valuation. If you're reading this, you’re already ahead. Stay there with our newsletter .