Why America Just Locked Away Its Best AI

Military personnel monitoring multiple screens in a command center

The federal government just turned OpenAI’s newest GPT-5.6 AI model into a controlled national security asset — and it decides who gets in the door.

Story Snapshot

OpenAI’s GPT-5.6 launch is a US-only preview for a small group of “trusted partners” at the direct request of the federal government.^[3]
Two weeks earlier, the government ordered rival Anthropic to block all foreign users from its top Fable 5 and Mythos 5 models over security fears.^[4]
Officials say these frontier models can spot software weaknesses that hackers could abuse, raising new cyber and infrastructure risks.^[1]
President Trump signed an executive order that creates a voluntary national security review process for advanced AI before broad release.^[4]^[6]

Government Puts GPT-5.6 Behind A Controlled Access Gate

OpenAI’s new GPT-5.6 series is not rolling out like a normal tech product; it is entering the world through a government-shaped tunnel.^[3] The company briefed federal officials on the model’s powers before launch, then agreed to start with a limited preview for a small group of trusted partners whose names were shared with authorities.^[3] Current reporting says about twenty partners will get access at first, including one route through Amazon’s Bedrock cloud service, keeping this tool inside a tightly managed circle.^[5]

OpenAI’s own blog and public comments stress that this narrow rollout is a temporary step, not the long-term plan for how Americans and allies should access AI.^[3] Company leaders say they are “uncomfortable” with the process and do not think government access or approval should become the default rule for future models.^[3] They warn that strict limits also keep powerful tools away from cyber defenders and foreign partners who help protect shared networks, even as Washington says the goal is better security.^[3]

National Security Fears Drive Tight Controls On Frontier Models

The Trump administration’s move on GPT-5.6 comes after growing concern that the most advanced frontier models can find software holes faster than human hackers or security teams.^[1] Both Anthropic’s Mythos series and OpenAI’s GPT-5.6 are reported to have an unprecedented ability to identify weaknesses in code that attackers can exploit in critical systems.^[1] That possibility has spooked investors and world leaders and pushed AI security from a niche topic into front-page news and cabinet-level briefings.^[6]

Two weeks before OpenAI’s US-only preview, the Department of Commerce ordered Anthropic to suspend foreign national access to its Fable 5 and Mythos 5 models based on national security concerns.^[4]^[15] Officials acted after learning about a jailbreak technique that could bypass guardrails and coax the model into listing cybersecurity vulnerabilities tied to infrastructure and other sensitive operations.^[10] Anthropic pushed back, saying the exploit seemed narrow and not universal, and questioned whether recalling a widely used model was justified without fuller evidence.^[10]

Trump’s Executive Order Aims For Security Without A Licensing Regime

Earlier in June, President Donald Trump signed Executive Order 14409, “Promoting Advanced Artificial Intelligence Innovation and Security,” to shape how Washington works with private AI developers.^[6] The order directs agencies to build a classified benchmarking system to judge the cyber capabilities of new models and decide when a system becomes a “covered frontier model.”^[6] For those frontier systems, companies can voluntarily give the government short-term access, up to thirty days, before wider release so officials can study risks and pick trusted early partners.^[6]

The executive order is careful to say it does not create a mandatory licensing or preclearance regime for AI models.^[6] It instead lays out a voluntary framework that encourages developers to engage federal experts, share access for a limited window, and collaborate on selecting early users who will help secure critical infrastructure.^[6]^[13] At the same time, it tells the Attorney General to prioritize criminal enforcement against anyone who uses AI to break into computers or carry out other online crimes, tightening the law enforcement focus on AI-enabled attacks.^[6]

Balancing Liberty, Innovation, And Real Cyber Risk

For conservatives who care about limited government, strong national defense, and a free market, this new AI playbook cuts both ways. On one hand, there is a clear and rising threat: frontier models like GPT-5.6 and Mythos can help spot flaws in software that hostile states, cybercriminals, or terrorists might use against American power grids, hospitals, or financial systems.^[1] Ignoring that risk would be reckless, especially when rogue actors are racing to weaponize AI for hacking and digital warfare.^[6]

Something historic just happened in AI governance and most people are framing it wrong.

The Trump administration asked OpenAI to restrict GPT-5.6 to a limited partner preview before broader release. Howard Lutnick personally warned Sam Altman not to launch without agency…

— Prasenjit Sarkar (@stretchcloud) June 26, 2026

On the other hand, Americans know how quickly “temporary” controls can turn into permanent bureaucratic choke points on innovation and speech. Critics already describe US-only previews and foreign bans as government overreach and anti-competitive, warning that Washington is now effectively deciding who gets the best tools in the AI race.^[1] OpenAI’s discomfort, Anthropic’s resistance, and the lack of public technical evidence about the claimed exploits all suggest that this new security line must be watched closely so it protects the country without quietly eroding freedom or fair competition.^[3]^[10]