The Trump administration has reportedly asked OpenAI to limit the distribution of its forthcoming model, GPT-5.6, to a select group of government-approved partners, citing potential security risks. The request from the White House's Office of the National Cyber Director and Office of Science and Technology Policy comes as the administration creates a framework for testing and evaluating the security of new models. OpenAI CEO Sam Altman confirmed the plans for a limited rollout in a memo to staff, stating that this was not the company's preferred long-term model but a necessary step while broader cyber policy frameworks are shaped.

The government's intervention was driven by GPT-5.6's "Mythos-like" capabilities, with a focus on ensuring companies implement adequate safeguards for highly advanced AI models. This follows the administration's clash with Anthropic, which disabled access to its Fable 5 and Mythos 5 models after being directed to block foreign nationals from using them. White House AI and Crypto Czar David Sacks expressed concern that Anthropic's AI models could be exploited as cyberweapons due to advanced cyber capabilities and reported jailbreak vulnerabilities.

OpenAI is introducing three tiers of GPT-5.6—Sol, Terra, and Luna—with the initial rollout restricted to a small group of vetted partners. Access is being delivered first through the API and Codex. At the top end, GPT-5.6 Sol is positioned as the flagship model, adding new reasoning controls like a "max effort" setting and an "ultra" mode that can deploy subagents for more complex workflows. The company highlighted strong benchmark results across coding, biology, and cybersecurity tasks, including performance on Terminal-Bench 2.1 and GeneBench v1.

Security and Safety Features

On the security front, OpenAI says Sol is better at spotting and fixing software vulnerabilities than carrying out full exploit chains. In testing across browsers like Chromium and Firefox, it was able to surface bugs and exploit components but did not independently produce end-to-end working exploits under the conditions evaluated. OpenAI outlined a layered safety system combining training, real-time monitoring during outputs, account-level signals, and tiered access controls. During the preview, some prompts may be slowed or blocked for extra review as the company fine-tunes false positives ahead of wider release.

Testing and Availability

Behind the scenes, OpenAI has logged more than 700,000 A100-equivalent GPU hours on automated red-teaming to probe jailbreaks, alongside external expert testing. The company described a rapid patch cycle to reproduce newly discovered jailbreak methods and fold fixes into future evaluations. Separately, GPT-5.6 Sol is slated to come to Cerebras in July, with throughput of up to 750 tokens per second, though access will start limited as capacity ramps. After the preview phase, OpenAI expects broader availability through ChatGPT, Codex, and its API.

OpenAI is reportedly leaning toward delaying its initial public offering (IPO) until 2027 as it pursues a potential $1 trillion valuation, according to a New York Times report citing three people involved in the deliberations. Advisers have presented CEO Sam Altman with two options: wait until 2027 for the trillion-dollar valuation or settle for a lower figure and list in 2026. Altman reportedly called any reduction a "non-starter." CFO Sarah Friar is said to be pushing for the delay, citing roughly $600 billion in compute infrastructure commitments through 2030 as a reason the company needs more runway before facing public-company reporting obligations.

Financial Burn and Valuation Skepticism

NYU professor and AI skeptic Gary Marcus stated there is "no rational argument" for buying a share of OpenAI at a trillion-dollar valuation. Marcus noted the company is losing roughly $2 billion a month on operations. The Wall Street Journal reported earlier this week that OpenAI has missed recent internal revenue targets, lending weight to skepticism about the company's financial trajectory. Previous reports indicated OpenAI spent about $3.7 billion in the first quarter of 2026 against revenue of about $5.7 billion.

Market and Competitive Context

The advisers' caution may also be tied to the rocky debut of SpaceX Inc. (NASDAQ: SPCX), whose shares have tumbled from above $225 to around $153 since its record IPO earlier this month. While the stock still trades above its $135 offer price, the roughly 30% retracement from its peak may have spooked executives weighing OpenAI's listing window. Competition is also intensifying, with Anthropic pursuing its own IPO plans after filing a confidential Form S-1 on June 1. Polymarket traders place OpenAI third in the race to have the best AI model by the end of 2026, at 11%, far behind Anthropic at 67% and Google at 13%.

Prediction Markets

Polymarket data indicates a shifting timeline for the listing. The market currently assigns a 78% probability that OpenAI will not list before December 31, 2026. A separate market gives Anthropic a 77% probability of beating OpenAI to the public markets.