GPT-6 Astra Ships, and OpenAI Says the Quiet Part Out Loud
OpenAI released GPT-6 Astra on September 3, and Greg Brockman didn't hedge: he called it a generational leap and said it could eventually be seen as the arrival of AGI. Labs usually let other people say that word. This time the president of OpenAI said it on launch day.
The pitch is agentic through and through: advances in coding, scientific research, cybersecurity, and computer use, tuned for professional environments where the model automates complex digital tasks end to end. It rolls out in phases to ChatGPT Plus, Pro, Business and Enterprise, plus the API and, notably, Amazon Web Services.
The most interesting part is the warning label. Astra is the first model OpenAI has designated as Critical on the cybersecurity axis of its Preparedness Framework, meaning it can find and exploit previously unknown vulnerabilities in well-protected systems without a human guiding each step. That's why companies in OpenAI's application-based cybersecurity program get access first. We covered the buildup in August when OpenAI admitted it had paused a model over exactly this threshold (https://clauday.com/article/8cfb7ba6-2901-4655-8374-be9ac06df3f5). The pause didn't end with a retreat, it ended with a guest list.
Which makes it three for three: Anthropic put Mythos behind vetted access, Google put Gemini Flash Cyber behind the Fairwind program, and now OpenAI ships its Critical-rated model to approved defenders first. Within one week, gating dangerous capability behind an application form went from an experiment to the industry standard. The system card is at https://deploymentsafety.openai.com/gpt-6-astra and the announcement at https://openai.com/index/gpt-6-astra/.
← Back to all articles
The pitch is agentic through and through: advances in coding, scientific research, cybersecurity, and computer use, tuned for professional environments where the model automates complex digital tasks end to end. It rolls out in phases to ChatGPT Plus, Pro, Business and Enterprise, plus the API and, notably, Amazon Web Services.
The most interesting part is the warning label. Astra is the first model OpenAI has designated as Critical on the cybersecurity axis of its Preparedness Framework, meaning it can find and exploit previously unknown vulnerabilities in well-protected systems without a human guiding each step. That's why companies in OpenAI's application-based cybersecurity program get access first. We covered the buildup in August when OpenAI admitted it had paused a model over exactly this threshold (https://clauday.com/article/8cfb7ba6-2901-4655-8374-be9ac06df3f5). The pause didn't end with a retreat, it ended with a guest list.
Which makes it three for three: Anthropic put Mythos behind vetted access, Google put Gemini Flash Cyber behind the Fairwind program, and now OpenAI ships its Critical-rated model to approved defenders first. Within one week, gating dangerous capability behind an application form went from an experiment to the industry standard. The system card is at https://deploymentsafety.openai.com/gpt-6-astra and the announcement at https://openai.com/index/gpt-6-astra/.
Comments