OpenAI's latest flagship model GPT-5.6 (Sol) was "held back" by the U.S. government for 12 days between the completion of training and public release; Sam Altman stated on CNBC that the process involved multiple rounds of discussions with officials including Secretary of Commerce Lutnick, Secretary of the Treasury Bessent, and National Cyber Director Cairncross, but OpenAI refused to disclose specific details.
Sam Altman: Review involved multiple rounds of talks with Commerce Secretary, Treasury Secretary, and National Cyber Director
According to reports, Sam Altman said on CNBC that the government review process for GPT-5.6 involved multiple rounds of discussions with officials including Secretary of Commerce Lutnick, Secretary of the Treasury Bessent, and National Cyber Director Cairncross; the AI Standards and Innovation Center (CAISI) under the Department of Commerce is currently leading the evaluation.
However, OpenAI declined to reveal who was involved in specific testing, what standards were used, or the basis for reviewing an AI model that is critical for public use, leaving external parties completely in the dark.
CAISI leads evaluation, executive order requires six cabinet agencies to establish review processes by early August
According to reports, an executive order issued after weeks of internal debate mandates that six cabinet agencies must establish formal AI model review procedures by early August 2026; this suggests that the review process GPT-5.6 has undergone is at best a "temporary patch" rather than an official system.
Sriram Krishnan, a senior former White House AI advisor, told the Financial Times, "There won't be an AI version of the FDA." Andy Konwinski, co-founder of Databricks and Perplexity, argued on X that there is no answer to who has the authority to approve models for deployment or to oversee them; he advocates establishing an "open consensus" model similar to the FDA, NIH, and national laboratories, allowing safety, interpretability researchers, and data experts to genuinely participate in review decisions.
Sol safety card cites external evaluations, internal government review remains opaque
According to OpenAI's disclosed content, GPT-5.6 Sol's Safety Card references external evaluations from three third-party organizations: the UK AI Safety Institute (UK AISI), the biosecurity organization SecureBio, and red team testing firm Irregular.
However, the actual review conversations and specific assessment methods conducted within the government are entirely unknown to the public; Dean Ball, a former Trump policy advisor now at OpenAI, suggested in an email that future oversight should be handled by third-party audit organizations approved by the government to establish a more transparent accountability mechanism.
Altman reportedly donates 5% stake to government fund, Greg Brockman is Trump’s biggest political donor
According to reports, political connections raise questions about potential conflicts of interest in GPT-5.6 review: Sam Altman has reportedly expressed willingness to donate up to 5% of OpenAI shares to a government fund for a "Trump account"; OpenAI President Greg Brockman is said to be the largest known political donor supporting Trump’s midterm election efforts.
David Siegel, founder of Two Sigma, proposed a hypothetical scenario at the Open Frontier seminar: a few companies control key technologies, the government evaluates these in secret labs, and the public and scientific community cannot access the decision-making process—an analogy that some interpret as reflective of current realities.
Frequently Asked Questions
How long was GPT-5.6 Sol held back by the U.S. government before official launch?
Reports indicate that GPT-5.6 Sol was "held back" by the U.S. government for 12 days between training completion and public release; Sam Altman stated on CNBC that the review involved multiple rounds of discussions with officials including the Secretary of Commerce, Secretary of the Treasury, and National Cyber Director, but OpenAI refused to disclose specific review standards and methods.
Which U.S. government agency is currently responsible for reviewing AI models?
According to reports, the Department of Commerce's CAISI (AI Standards and Innovation Center) is currently leading evaluations; however, an executive order requires six cabinet agencies to establish formal review processes by early August 2026, meaning the current mechanism is temporary and a formalized review system has yet to be implemented.
What third-party organizations did OpenAI disclose regarding GPT-5.6's safety evaluation?
OpenAI's publicly released Sol Safety Card cites external evaluations from the UK AI Safety Institute (UK AISI), biosecurity organization SecureBio, and red team testing firm Irregular; details of internal government review conversations are not disclosed.
