The Trump administration is set to issue an executive order this week that will establish a framework for AI labs to share new models with the US government up to 90 days prior to their public release. This action follows the unveiling of advanced AI systems, such as Anthropic’s Mythos Preview and OpenAI’s GPT-5.5, which raised concerns within national security circles due to their potential to exploit vulnerabilities in critical infrastructure. The NSA is designated to conduct classified evaluations of these models, but there is internal conflict over the extent of oversight, with some officials advocating for stringent safety measures while others warn that excessive regulation could hinder AI innovation. The framework’s voluntary nature raises questions about its implications for government reach into private tech development, particularly given the dual-use nature of these frontier systems.

NSA: The National Security Agency (NSA) is a U.S. intelligence agency responsible for signals intelligence and cybersecurity, including protecting federal networks and monitoring foreign cyber threats. In this story, the NSA is slated to handle classified testing of powerful new AI models before their public release, positioning the agency as a central evaluator of potential security risks posed by advanced AI capabilities.
Axios: Axios is a U.S.-based digital media outlet that focuses on concise, policy-centric and technology-focused news coverage for professional and political audiences. In this context, Axios is one of the outlets reporting that national security officials were unnerved by the latest AI model capabilities and advocating for an executive order granting early government access to such systems.
OpenAI: OpenAI is an artificial intelligence research and deployment company known for building state-of-the-art language and multimodal models used by consumers, enterprises, and developers worldwide. In this news, OpenAI’s GPT-5.5 model demonstrated advanced cyber-relevant capabilities that, alongside Anthropic’s work, alarmed U.S. national security officials and motivated the push for government preview access to future frontier models.
Nextgov: Nextgov is a media publication that covers technology, cybersecurity, and digital policy in the U.S. federal government, with a particular focus on how agencies adopt and regulate emerging technologies. In this story, Nextgov is cited as a source detailing how parts of the national security and federal technology community reacted to frontier AI model releases and pushed for structured pre-release evaluations by government agencies.
Anthropic: Anthropic is an AI research company that develops large language models and emphasizes safety, alignment, and controls around how its systems are deployed. Here, the release of Anthropic’s Mythos Preview model, which reportedly showed strong capabilities in probing computer networks and critical infrastructure, helped trigger national security concerns that led the administration to pursue a pre-release review framework.
Trump administration: The Trump administration refers to the current U.S. executive branch under President Donald Trump, which has taken an increasingly hands-on approach to regulating advanced AI systems in the context of national security and critical infrastructure risks. In this news, the administration is preparing an executive order that would establish a framework for AI labs to give the government early access to new frontier models, reflecting internal debates over how aggressively to oversee private AI development.

AI_governance_debate: Within the U.S. policy community there is an active split between officials who prioritize stringent safety evaluations for frontier models and those who warn that heavy-handed controls could push AI innovation and research activity overseas.
National_security_focus: Recent policy discussions in Washington have increasingly framed frontier AI systems as dual-use technologies whose offensive cyber and surveillance capabilities warrant closer intelligence-community oversight.
Industry_government_relations: Major AI labs have been publicly emphasizing cooperation with governments on safety and security testing, even as they raise concerns in interviews and policy forums about maintaining competitive secrecy around unreleased model capabilities.