US tightens AI export controls, extending restrictions to OpenAI's GPT-5.6

The US government has extended export controls to OpenAI's next-generation AI model, following similar restrictions already imposed on Anthropic's Mithos.

As concerns grow over AI-driven cybersecurity threats, competition among nations over AI security sovereignty is intensifying. South Korea's government is racing to secure access to the latest global AI models but has yet to achieve any clear results.

According to industry sources Monday, OpenAI announced June 26 on its website the launch of GPT-5.6 Sol — described as carrying the most powerful security stack in the company's history. The company disclosed that it had conducted "a limited early release to a small number of trusted partners at the request of the US government" and that "the list of participating partners has been shared with the US government."

The move follows an executive order signed by US President Donald Trump earlier this month requiring government approval before AI models can be deployed, part of a broader tightening of export controls on AI.

OpenAI Korea clarified that the disclosure meant the company had shared a list of participating organizations in accordance with US government approval procedures, and did not indicate that those organizations had been granted access to GPT-5.6 Sol.

With early access now restricted for both Anthropic's Mithos and OpenAI's latest model, concerns are mounting that South Korean companies could face setbacks in their cybersecurity preparedness.

The US Commerce Department earlier announced export controls on the Mithos 5 model. Reports have since emerged that some restrictions have been eased, but access for South Korean firms — including SK Telecom, Samsung Electronics and SK hynix — is still expected to remain limited. The US government has confined use to specific domestic companies and institutions it has vetted as secure.

That restriction also explains why South Korea's Ministry of Science and ICT, despite joining OpenAI's Government and Trusted Access Community program on May 26, has been unable to secure access to GPT-5.6 Sol.

The government is struggling to keep pace. As the release cycle for high-performance AI models shortens and concerns over AI-driven cyber threats mount, competition over AI security sovereignty is growing more acute. Experts say South Korea must pursue global cybersecurity cooperation while also building its own sovereign AI capabilities.

A Ministry of Science and ICT official said the ministry is following procedures required under the US executive order and is "in ongoing discussions with both OpenAI and Anthropic regarding access to next-generation AI models."

Meanwhile, concerns over AI security have consistently grown since Anthropic unveiled its next-generation autonomous AI model, Mithos, in April.

A cybersecurity threat outlook survey released by Samsung SDS — drawing on responses from 667 domestic IT and security professionals — found that AI-based security threats topped the list of cybersecurity risks expected to have the greatest impact on businesses this year, cited by 81.2 percent of respondents.

Respondents expressed particular concern that AI could expose sensitive information through interactions involving humans, AI agents and agentic AI systems.