The Trump administration has pivoted from a to active oversight, now pressuring to delay its latest AI model release due to escalating safety fears. This shift follows a precedent where the White House forced to disable its model after researchers identified posing bio-weapon and cyberattack risks. The story underscores a growing tension between the government’s drive for national security control and AI companies’ push for , as highlighted by ’s recent policy paper diverging from White House safety benchmarks.
OpenAI announced it is limiting the release of its new GPT-5.6 models (Sol, , ) to a small group of trusted partners whose participation has been shared with the U.S. government, complying with a formal request from the White House due to safety concerns. The company stated this preview is a short-term step toward broader availability in coming weeks and explicitly criticized the government access process as not suitable for becoming the long-term default.
Initial access to the GPT-5.6 models (Sol, Terra, Luna) was extended to exactly 20 trusted partners. This small group includes routes like Amazon's Bedrock platform, though OpenAI did not disclose the specific names of the approved partners.
OpenAI stated this limited preview is a short-term step, with broader public availability expected in the coming weeks. The company is working toward a phased rollout to make the models generally available shortly after this government-approved phase.
OpenAI explicitly criticized the government access process as unsuitable for becoming the long-term default. The company argues that restricting access keeps the best tools away from users, developers, enterprises, cyber defenders, and global partners who need them.
Following the Anthropic precedent and amid policy divergence, the White House formally asked OpenAI to slow the release of its new GPT-5.6 model over safety concerns. The and requested limited initial access, marking the second major intervention in AI governance.
The White House formally requested OpenAI to slow the release of its new GPT-5.6 model due to safety concerns. This intervention, led by the Office of the National Cyber Director and Office of Science and Technology Policy, marks the second major government action in AI governance following a similar precedent with Anthropic.
The government requested limited initial access for GPT-5.6 over escalating safety fears, specifically regarding cybersecurity and bio-weapon risks. This shift from a laissez-faire approach to active oversight aims to prevent the model from autonomously generating malicious code or exploiting software vulnerabilities before robust safeguards are confirmed.
OpenAI released a policy paper calling for civilian agencies () to oversee frontier AI safety, diverging from the White House's plan to place the NSA in charge. The paper criticized the administration's proposal for a , highlighting a clash over who controls safety standards.
The paper criticized the administration's proposal for a classified benchmarking process, highlighting a clash over who controls safety standards between civilian oversight and national security control.
President Trump signed an titled 'Promoting Advanced Artificial Intelligence Innovation and Security,' directing federal agencies to establish a framework for secure deployment of . The order created a voluntary process for developers to submit models for government testing up to 30 days before public release.
The Trump administration abruptly imposed on Anthropic, forcing the company to disable its Fable 5 and . This was the first time the government told a leading AI company its models were unsafe for public consumption, establishing a precedent for future interventions.
The Trump administration imposed foreign export restrictions citing national security authorities due to a reported AI jailbreak, forcing Anthropic to disable Fable 5 and Mythos 5 for every customer worldwide on June 12, 2026.
Security researchers found a way to bypass guardrails in Anthropic's , posing cybersecurity risks for bio-weapons and cyberattacks. Amazon CEO raised concerns about these vulnerabilities, prompting urgent government attention to the model's safety profile.
Researchers used Unicode/homoglyph tricks, long-context framing, and multi-agent prompting to manipulate the model's interpretation of context, bypassing safety checks and obtaining step-by-step stack overflow exploit guidance.
The bypassed guardrails pose cybersecurity risks for bio-weapons and cyberattacks, as the model can generate outputs normally blocked, including exploit development and vulnerability reproduction.
White House representatives shared oversight proposals with executives from , Google, and OpenAI, signaling a significant shift toward governmental supervision. These meetings marked the first direct engagement between the administration and major AI firms regarding federal review protocols.
These meetings were the first direct engagement between the Trump administration and major AI firms regarding federal review protocols. They signaled a decisive shift from the previous hands-off approach to active governmental supervision of AI model releases.
By May 2026, the administration shifted course from its hands-off approach, discussing an executive order to create an uniting industry and government. The proposal explored oversight mechanisms including of new models before public release.
The proposal explored systematic government review of new AI models before they are released to the public. This would create an AI task force uniting industry and government, signaling a move toward active oversight rather than the previous laissez-faire stance.
The White House issued an , setting the stage for federal control over safety standards. This move centralized oversight authority and signaled the administration's intent to establish a unified national approach to AI safety.
President Trump signed the executive order titled 'Ensuring a National Policy Framework for Artificial Intelligence' on December 11, 2025. It directs federal agencies to challenge state AI laws deemed overly burdensome or ideologically biased within 30 days.
The order establishes a DOJ AI Litigation Task Force to challenge state laws within 30 days and restricts Broadband Equity, Access, and Deployment funds for states with 'onerous' laws. It also mandates the Commerce Department to identify unconstitutional state laws by March 11, 2026.
Upon returning to office, President Trump quickly dismantled the requiring AI developers to conduct safety assessments. The administration advocated for a to foster global competition against China, marking an initial shift toward nonintervention in AI governance.
President Trump revoked Biden's Executive Order on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence just hours after taking office in January 2025. This repeal also included instructions to review a National Security Memo governing national security AI use, effectively removing safety assessments for developers.
The administration adopted a laissez-faire stance to foster global competition against China, viewing safety assessments as barriers to American AI leadership. This marked an initial shift toward nonintervention, aiming to accelerate unchecked AI deployment across American life.