June 27, 2026, 09:17June 27, 2026, 09:17
ChatGPT developer OpenAI is restricting access to its latest AI model. Only a coordinated small group of trusted partners will have access to the preview version of models in the GPT 5.6 series, OpenAI said.
The restricted publication was requested by the government. OpenAI said the approvals were initially only granted in the USA. They are working on giving partners from other countries access, and this could happen as early as next week.
OpenAI is restricting access to its latest AI model.Image: keystone
At the same time, OpenAI said that it does not believe that such access controls by the authorities will become the norm in the long term. At the moment, agreeing to this is seen as the best way to expand access in the coming weeks.
This statement can be understood as an indication of problems at rival Anthropic. The company initially released its new, most powerful model – and then had to withdraw it a few days later at the request of the US government.
Fear of cyberattacks with AI help
The trigger was warnings that precautions against misuse of the software could potentially be undermined. Anthropic’s new AI model is particularly good at finding software vulnerabilities. This makes programs more secure – but in the wrong hands, such AI could be used as a cyber weapon. Anthropic blocks the corresponding abilities in the publicly available version. However, the US government demanded that only American companies and individuals be allowed to have access to the AI model.
According to the company, the most powerful version of the new OpenAI model GPT-5.6, called Sol, can independently carry out tasks in programming, biology and cybersecurity. OpenAI emphasized that protective measures had been taken to prevent misuse. At the same time, evaluations cannot take every multi-stage attack into account, the company warned. (hkl/sda/dpa)