OpenAI, the creator of ChatGPT, announced on Friday that it is limiting the rollout of its latest artificial intelligence model following a request from the U.S. government led by President Donald Trump. This move is part of a unique initiative by the government to evaluate AI technologies that could potentially pose cybersecurity threats.
The new AI model, named GPT-5.6 Sol, will initially be accessible only to a select group of trusted partners approved by the Trump administration, as confirmed by OpenAI. The company emphasized that it does not consider this government oversight process as a permanent arrangement. Instead, it views this as a temporary measure on the path towards wider availability in the near future.
In a parallel development, OpenAI’s decision to introduce its advanced AI system in stages mirrors recent government actions targeting Anthropic, a competitor renowned for developing the Claude chatbot. Anthropic promptly withdrew its latest AI models, Fable 5 and Mythos 5, shortly after their public release to adhere to a directive from the Trump administration prohibiting their use by foreigners.
The White House expressed continued collaboration with leading AI research labs to address the complexities of scaling this rapidly evolving technology. Concerns escalated after Anthropic cautioned earlier this year that its Mythos model had the capability to identify software vulnerabilities that could be exploited by malicious actors, potentially jeopardizing critical computer networks globally.
The executive order signed by President Trump in June on AI oversight laid down a framework for assessing the national security implications of cutting-edge AI systems for up to 30 days before their public launch. While participation by AI developers is voluntary, the framework’s full implementation is pending.
OpenAI, headquartered in San Francisco, stated that its new Sol model excels in aiding in the identification and resolution of vulnerabilities rather than executing cyberattacks, staying within the company’s risk tolerance level. Nonetheless, the company acknowledged the existence of unforeseen risks, particularly when integrating its model with other tools.
The company emphasized the need for caution due to the model’s enhanced capabilities and potential combined risks. This cautious approach led to the phased release of the AI model to mitigate any potential threats.
Various cybersecurity experts have criticized the government’s actions, which resulted in Anthropic halting the operations of Fable, a variant of the Mythos model marketed as a safer alternative. The cybersecurity community, including Alex Stamos, a cybersecurity expert from Stanford University, expressed skepticism about the factual basis for this action, highlighting that similar risks are present in publicly available AI models, including those originating from China.
Stamos, also the chief product officer at AI security firm Corridor and a former chief security officer at Meta, emphasized that the government’s actions might hinder the U.S.’s competitiveness against China in the AI domain. The expert emphasized the importance of strategic decisions to maintain leadership in this technological race.