OpenAI announced the release of its most powerful model family, GPT-5.6 (Sol, Terra, and Luna), on July 9, 2026, marking the end of its limited access period. This launch represents the first public model release ever coordinated with the Trump administration, beginning with a preview for select partners on June 26 under a new policy framework for cooperation between AI labs and the government. Approximately 20 organizations were granted initial access before the model was made available to the general public.
The White House voiced concerns regarding potential cyber threats that could emerge from the widespread distribution of such sophisticated models. Consequently, OpenAI is voluntarily implementing a phased rollout, providing access first to government-approved partners before a planned expansion. This decision highlights a shift beyond mere architectural breakthroughs toward a new era of strategic safety collaboration between the private sector and the state—a move that will define industry standards for years to come.
This development traces back to a landmark move by President Trump, who signed an executive order on June 2, 2026, establishing a voluntary framework for federal agencies to test new AI models prior to their public release. Companies can choose whether to subject their models to these additional government reviews. OpenAI opted to participate, engaging in a joint analysis period with the White House and various agencies that lasted over a month. A pivotal moment occurred on June 3 in Washington, when CEO Sam Altman met with administration officials and members of Congress, including Speaker Mike Johnson and Minority Leader Hakeem Jeffries, to engage with stakeholders from across the political spectrum.
However, OpenAI’s trajectory stands in stark contrast to the challenges faced by Anthropic. During the same week OpenAI announced its limited release, Anthropic launched its two most powerful models: Fable 5, which saw its first public release on June 9, and Mythos 5, intended for a select group of partners. Just three days later, on June 12, the government intervened: Commerce Secretary Howard Lutnick issued a directive citing national security concerns, ordering Anthropic to immediately terminate access to both models for all foreign nationals, including the company’s own employees.
The cause was a report indicating that Amazon researchers had discovered a way to bypass Fable 5’s safeguards to access the full cyber capabilities of Mythos. Because Anthropic was technically unable to segment users by nationality, the company was forced to shut down the models for its entire global user base. This represented one of the most aggressive uses of export controls ever applied to a deployed commercial AI model.
These restrictions lasted nearly three weeks, testing the company’s resilience. Following intensive negotiations with the Cabinet, Lutnick authorized a partial restoration on June 26, allowing Mythos 5 to be used by roughly one hundred vetted American organizations and federal agencies. On June 30, export controls were lifted entirely, and Fable 5 returned to global networks on July 1. Nevertheless, the cost was significant—Chinese developers gained valuable time to advance, and industry-wide anxiety has noticeably intensified.
The disparate treatment of the two companies reveals the complex dynamics of state policy in the era of frontier models. By coordinating its plans with the White House in advance, OpenAI avoided heavy-handed sanctions. Conversely, Anthropic was hit with a sudden ban, partly due to its lesser political influence and partly because its own decision to market Mythos as "too dangerous" for general release provided officials with a pretext for action. While critics highlight the inconsistent application of rules, national security advocates maintain that both models posed genuine high-level cyber risks.
For developers and researchers, this represents a turning point where frontier models must now navigate informal government reviews in addition to internal testing before a wide release. While this may dampen the global pace of innovation, it simultaneously bolsters confidence in market-ready models among corporate and government sectors—or at least creates that impression.
Serious questions remain, however, regarding the long-term viability of this approach. OpenAI has clearly signaled that it does not view government approval as a permanent standard, despite participating in this voluntary model. Anthropic and other labs remain in a state of uncertainty, wondering if such restrictions will continue, under what conditions, and based on what criteria?
There is a total absence of independent verification for these government-led tests; all safety claims rely solely on the positions of the companies and officials involved, without public technical reports or third-party audits. Secretary Lutnick even reserved the right to revisit these decisions at any time "should circumstances change."
Ultimately, the July release of GPT-5.6 demonstrates that by 2026, the speed and conditions for bringing powerful AI to market are determined as much by political relationships as by technical readiness. This is redefining the competitive landscape and setting a precedent that will impact every future frontier model release. The events of June will be remembered for a long time.
