The US government is now handpicking who gets OpenAI’s most powerful new AI, raising big questions about security, freedom, and who really controls the future of technology.
Story Snapshot
- OpenAI’s new GPT-5.6 model is in a US-only preview, limited to a small list of “trusted partners” chosen with Washington’s input.[1][3]
- Two weeks earlier, the government ordered rival Anthropic to block all foreign users from its top models over national security fears.[4][15]
- Officials say these frontier models can spot software weaknesses that hackers and hostile regimes could exploit.[1][4]
- President Trump signed a voluntary review order that aims to secure powerful AI without creating a permanent licensing regime.[13][14]
New GPT-5.6 rollout puts Washington in the driver’s seat
OpenAI has launched a preview of its new GPT-5.6 series, but only for partners in the United States and only after briefing the federal government first.[1][3] The models, named Sol, Terra, and Luna, are described as OpenAI’s latest “frontier” systems with stronger intelligence and safety guardrails than past releases.[4][6] Access is starting with a tight circle of about twenty trusted entities, including one pathway through Amazon’s Bedrock platform, and those partner identities have been shared with authorities.[5] OpenAI says this restricted launch is a temporary step before a wider public release in coming weeks.[3][6]
OpenAI’s own statement makes clear this process is not the company’s ideal long-term approach and that it is “uncomfortable” with government-directed access, but it is complying under an existing Defense Department agreement that already lets the Pentagon use its models.[3][6] This agreement and the current preview show how closely frontier AI and national defense are now linked. For conservative readers, that tight link cuts both ways: it can help keep America ahead of China and Russia, but it also raises the risk of Washington using security as a reason to control powerful tools that can boost free enterprise, local innovation, and individual productivity.[12][14]
Security fears after Anthropic is ordered to cut off foreign users
The OpenAI rollout comes just two weeks after a shock order against Anthropic, another leading AI firm.[1][4] The Department of Commerce told Anthropic to suspend access to its most advanced Fable 5 and Mythos 5 models for all foreign nationals, citing national security worries about a “jailbreak” that could bypass safety guardrails and expose cyber vulnerabilities in critical systems.[10][15] Anthropic responded by disabling those frontier models for every customer while it challenged the directive, arguing that the evidence pointed to a narrow, limited exploit rather than a broad systemic failure.[10][15] That clash showed the growing tension between federal agencies demanding tighter controls and private developers who say broad shutdowns are not justified by the technical facts.
Reports on both Anthropic’s Mythos and OpenAI’s GPT-5.6 stress their “unprecedented” ability to find weaknesses in software that hackers could use to break into networks, including infrastructure and other high-value targets.[1][4] Yet, so far, there is no public technical report, no catalog of specific vulnerabilities, and no independent audit released to back those sweeping claims.[1][4] From a rule-of-law and transparency perspective, this is troubling. When government uses emergency powers that affect global access and billions of dollars in innovation, citizens expect clear, verifiable evidence, not just classified briefings and unnamed concerns. Without that sunlight, it becomes harder to distinguish honest security work from simple bureaucratic overreach.
Trump’s executive order aims for security without permanent licensing
Earlier this month, President Donald Trump signed an executive order titled “Promoting Advanced Artificial Intelligence Innovation and Security,” laying out how Washington will handle frontier AI models going forward.[13][14] The order directs agencies to build a classified benchmarking system that scores advanced AI models on cyber capabilities and decides when a system counts as a “covered frontier model.”[13][14] For those covered models, the developer can voluntarily give the government up to thirty days of early access before broader release so officials can test for risks and help pick trusted partners for initial deployment.[13][14] Importantly for conservatives worried about a permanent licensing regime, the order explicitly forbids turning this into mandatory federal preclearance for any AI model.[13][14]
The order also tells the Attorney General to prioritize criminal enforcement against people who use AI to hack into computers, damage systems, or commit other crimes.[14] This focus on going after bad actors, rather than regulating tools themselves, aligns more closely with traditional conservative views: punish criminals, not technology. At the same time, the voluntary framework gives the national security community a formal path to ask for controlled rollouts, as happened with GPT-5.6 and Anthropic’s models.[1][3][13] That mix of freedom and caution depends on the government staying within the voluntary lane and avoiding backdoor pressure that effectively forces companies into compliance.
Balancing national defense with free markets and open innovation
Since 2024, the United States has steadily moved from focusing on hardware export controls, such as advanced chips, to tighter oversight of software model weights and access.[11][12] A previous “AI Diffusion Rule” that would have imposed sweeping controls on which countries could get advanced models was rescinded in 2025 after industry and allies warned it would choke American innovation and strain relationships abroad.[11][16] The current approach uses targeted directives, like the order against Anthropic, along with voluntary frameworks like Trump’s recent executive order, instead of blanket bans.[13][14] That shift is meant to protect national security while keeping US firms competitive in a fast-moving global race.
OpenAI just previewed GPT-5.6 Sol. And it's a big one.
Three models in the family:
→ Sol: flagship, max intelligence
→ Terra: balanced, 2x cheaper than GPT-5.5
→ Luna: fast and affordableWhat's new with Sol specifically:
A new ultra mode that spins up multiple subagents… pic.twitter.com/45ryBOnB24— Khalid (@KhalidDevLog) June 26, 2026
For everyday Americans, the stakes are high. On one side, hostile regimes and criminal gangs are eager to use frontier AI to probe our power grids, banks, and military networks. On the other, heavy-handed controls could let unelected bureaucrats decide who gets world-leading tools, freezing out small businesses, conservative states, and global allies who share our values. OpenAI itself warns that US-only limits keep these defensive tools away from foreign cyber partners who help find and fix vulnerabilities, which could actually weaken global security if misapplied.[1][3][6] The path forward needs what conservatives have always demanded: clear evidence, transparent rules, and firm limits on government power, even when the technology is new and the fear is real.
Sources:
[1] Web – OpenAI launches US-only preview of its new powerful AI model
[3] Web – OpenAI launches limited release of new model in US only
[4] Web – OpenAI limits release of new AI model amid US request
[5] Web – OpenAI limits release of new model under pressure from US
[6] Web – OpenAI limits release of new AI model amid US request By Investing.com
[10] Web – OpenAI restricts limited release of new model to US only
[11] Web – OpenAI launches limited release of new model in US only
[12] Web – Trump administration asks OpenAI to stagger release of new model …
[13] Web – OpenAI: Latest news and insights – Computerworld
[14] YouTube – OpenAI Just Unlocked the #1 Feature Businesses Begged For
[15] YouTube – OpenAI Just Dropped PRISM: Things Just Got Serious
[16] Web – Model Release Notes | OpenAI Help Center
