Directive that manages operations of a nation's executive branch
The White House has issued an executive order addressing cybersecurity risks posed by artificial intelligence, aiming for less-stringent industry oversight than previously planned. The action follows a quieter White House process and private signing by the president.
The White House has issued a scaled-back executive order asking frontier AI firms to voluntarily share advanced models 30 days before release for cybersecurity review. Last week the administration has imposed export-style restrictions that forced Anthropic to cut access to its Fable and Mythos models, prompting industry alarm about ad hoc controls and the limits of the voluntary framework.
A federal judge in Boston has voided the Trump administration's $100,000 fee on H‑1B visa petitions, ruling the payment functions as a tax that Congress did not authorize. The administration has filed a notice of appeal, and parallel lawsuits and appeals are proceeding in other federal courts, leaving the policy's fate to the appeals process.
Anthropic has said it has disabled access to its Fable 5 and Mythos 5 models after the U.S. Commerce Department has ordered the company to suspend foreign‑national access on national security grounds. Anthropic is complying while disputing the governments evidence of a narrow "jailbreak" and is working to restore access; other Anthropic models remain available.
The United States has tightened export controls on the most advanced AI models, restricting access for foreign nationals. Anthropic has widely released Fable, a limited version of Mythos, but access remains constrained due to cybersecurity concerns. Anthropic says the action is a misunderstanding and hopes to restore access soon; the Commerce Department has not commented. The move follows a presidential directive aimed at vetting national security risks in AI before public release.
Anthropic's Mythos model has identified vulnerabilities in highly secure U.S. government systems during tests run with U.S. intelligence agencies under Project Glasswing. Officials say findings show rapid detection, not immediate exploitation; collaboration aims to shore up cybersecurity as tensions with the administration grow.
OpenAI has previewed GPT-5.6 models, including Sol, Terra and Luna, to a select group of partners under government oversight. The move follows President Trump’s executive order prompting voluntary pre-release review of frontier AI. OpenAI says broader access will come in coming weeks, while stressing this approach is not a long-term default.
A Washington judge has ruled that the Defense Department’s media escort policy violates the First Amendment and has issued a preliminary injunction preventing its enforcement while The New York Times pursues its legal challenge. The Times, which has sued the department twice in five months, says the decision upholds press rights to cover Pentagon operations.