OpenAI Launches GPT-5.6 Under First Government-Gated AI Rollout in US History

Unprecedented Government Control Over AI Model Release

OpenAI unveiled its GPT-5.6 model suite on Friday. This launch marks the first government-gated frontier AI release in United States history under an unprecedented access framework. The company restricted initial access to roughly 20 organizations individually approved by the federal government. This move stems from a Trump administration executive order issued June 2, 2026, which requires federal agencies to benchmark and assess frontier AI models with advanced cyber capabilities before broad public release.

The rollout represents a dramatic shift in how cutting-edge artificial intelligence reaches the market. OpenAI signaled its dissatisfaction with the arrangement, stating that the restricted approach keeps the best AI tools from users, developers, enterprises, cyber defenders, and global partners who need them. The company wrote that it does not believe this kind of government access process should become the long-term default, though it acknowledged taking this short-term step as the strongest path to broader availability in the coming weeks.

The White House’s Office of the National Cyber Director and Office of Science and Technology Policy requested OpenAI limit access while a formal evaluation process targeting an August 2026 deadline is built out. All entities receiving initial access are US-based, though OpenAI expressed hope to add foreign partners next week. Employees of approved companies based abroad in supported countries, including the UK and Australia, will have access to the model.

Three-Tier Model Suite With Flagship Sol

GPT-5.6 ships in three distinct tiers designed for different use cases and budgets. Sol serves as the flagship model, engineered for hard reasoning, coding, and agentic work requiring sustained focus over long-horizon tasks. Terra functions as a mid-tier workhorse positioned as competitive with GPT-5.5 at half the cost for high-volume work. Luna offers the most affordable and fastest option, optimized for everyday tasks where speed matters more than maximum capability.

OpenAI says the suite excels particularly at coding, cybersecurity, and biology applications. Sol introduces two additional operational modes that expand its capabilities beyond standard inference. The max mode enables deeper reasoning for complex problems, while the ultra mode deploys sub-agents to decompose and tackle multifaceted tasks. The ultra mode evokes OpenClaw and may signal the influence of OpenClaw creator Peter Steinberger’s work at OpenAI.

Pricing reflects the tiered approach to accessibility and performance. Sol costs $5 per million input tokens and $30 per million output tokens, positioning it at nearly half the cost of Anthropic’s Claude Fable 5, which runs $10 input and $50 output. Terra comes in at half Sol’s price at $2.50/$15, while Luna undercuts Terra at less than half its cost with rates of $1/$6 per million tokens.

Cybersecurity Capabilities Trigger Government Scrutiny

The restricted rollout centers on Sol’s advanced cybersecurity capabilities, which sparked federal concern. The flagship model scored 96.7% on OpenAI’s internal Capture-The-Flag cybersecurity evaluations, crossing what the company classifies as a high cyber risk threshold. Despite this impressive performance, OpenAI maintains Sol did not cross a cyber critical threshold under its internal preparedness framework for measuring dangerous AI capabilities.

OpenAI emphasized that Sol demonstrates greater proficiency at helping defenders find and fix vulnerabilities than at executing end-to-end attacks. The model did not autonomously complete full-chain exploits in Chromium or Firefox testing, according to company assessments. The firm stated that Sol remains better suited for defensive cybersecurity work than offensive operations, a distinction critical to its safety evaluation.

“GPT-5.6 is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model,” OpenAI wrote in its announcement, appearing to reference recent jailbreaking challenges faced by rival Anthropic.

The company described Sol as having its most robust safety stack to date, with strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse. OpenAI worked with third-party testers during development and said these external evaluators will continue testing the model for the next two weeks. The preview period faces close monitoring by the Trump administration, with OpenAI noting that safeguards may occasionally intervene during this supervised release phase.

Echoes of Anthropic’s Mythos Controversy

The government-controlled release parallels the recent launch troubles of Anthropic’s Mythos model. Anthropic, OpenAI’s close rival, initially delayed widespread release of Mythos voluntarily due to its powerful cyber-hacking capabilities. The US government later ordered Anthropic to stop foreign nationals from accessing public versions of the model entirely. Anthropic ultimately pulled the technology completely under federal pressure, establishing a precedent that now appears to influence OpenAI’s rollout strategy.

The parallel situations highlight growing government concern over frontier AI models with advanced cyber capabilities reaching adversarial hands. Both companies faced similar federal scrutiny, though OpenAI appears to have preemptively negotiated its restricted release rather than facing a post-launch intervention. The coordination with government agencies before Friday’s announcement suggests OpenAI learned from Anthropic’s more contentious experience with federal regulators.

OpenAI’s public pushback against permanent government gatekeeping reflects broader industry tensions over AI regulation. The company maintains this represents a short-term compromise rather than an acceptable long-term model for innovation. As OpenAI works with the White House to develop vetting and deployment frameworks required under Trump’s executive order, the technology sector watches closely to see whether government-managed access becomes standard practice for frontier AI releases or remains an exceptional measure for models with elevated security implications.