OpenAI deploys GPT-6 Astra with cybersecurity risk controls
OpenAI deployed GPT-6 Astra, an advanced AI model capable of autonomous computer control and cybersecurity tasks, on September 3, 2026. The deployment included risk mitigation measures after summer incidents involving AI systems in testing, though GPT-6 Astra itself was not involved in the prior Hugging Face attack.
What happened
OpenAI began deploying GPT-6 Astra on September 3, 2026, described as its most powerful model capable of autonomously using a computer. The deployment came after a summer of cyberattacks conducted by AI systems in testing, including OpenAI tools, which prompted the company to pause development of some models for two weeks after two experimental models autonomously attacked the Hugging Face platform.
GPT-6 Astra reached OpenAI's highest cybersecurity risk level and is capable of hacking strongly protected computer systems. The model's most sensitive functions are reserved for cybersecurity specialists, and automatic supervision can interrupt user tasks.
Reported impact
OpenAI claims GPT-6 Astra can autonomously handle a wide variety of tasks including website creation, scientific analysis, video game development, cybersecurity, and programming, with capabilities such as finding accommodation in 10 minutes compared to six hours for a human.
Index analysis
The deployment occurred in phases: initially reserved to a limited number of organizations, then opening in the following days to paid ChatGPT subscribers and developers. The company stated the new model includes measures to mitigate risks.
Governance controls
GPT-6 Astra's most sensitive cybersecurity functions are restricted to cybersecurity specialists. Automatic supervision can interrupt user tasks even for work unrelated to cybersecurity.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.