OpenAI says the AGI era starts now, and it is betting a new model called GPT-6 Astra to prove it. The company rolled the model out on September 3, 2026, with a clear message and a clear plan. Greg Brockman, OpenAI’s president, told reporters the company sees this as a turning point. He ended a press briefing with four words: “Welcome to the AGI era.”
That line will travel. It is bold, and it lands right where the industry has been arguing for years. AGI has no single agreed test. It is a fuzzy target, not a finish line. Brockman knows that. He said it is not unreasonable to feel we are now in the AGI era, and he left the final judgment to users.
Here is what Astra is supposed to do. OpenAI calls it a generational leap for cybersecurity, software engineering, and office work. Brockman said as much on the call. “GPT-6 Astra is a ‘generational leap’ for cybersecurity, software engineering and office work.” The model is built to use a computer like a person would. It can work across browsers, spreadsheets, websites, and desktop apps. It can write, test, and debug code. It can create and manipulate documents and presentations. It can handle multi-step tasks without constant hand-holding.
It is also the first OpenAI model to hit the company’s critical cybersecurity capability threshold under its Preparedness Framework. That is a big deal inside the company. It means Astra is strong enough at cyber tasks that OpenAI had to pause and add stricter guardrails before shipping. The pause came after a security incident at Hugging Face, the platform where many AI models live and get shared. OpenAI said it delayed Astra’s development to improve safety tooling and to make sure the model could be monitored.
So Astra is arriving with limits. The version going to select enterprise customers in OpenAI’s Daybreak program can do common cyber defense work. Think vulnerability validation, malware analysis, and detection engineering. It will not let those customers develop exploits or run tasks that could be used to conduct cyber attacks. The wider rollout to ChatGPT Plus, Pro, Business, and Enterprise users will also refuse advanced cybersecurity tasks. OpenAI says it may run extra security checks that can slow, pause, or stop legitimate work, including defensive cybersecurity. That is the trade-off. Power with brakes.
Brockman also framed Astra as OpenAI’s attempt to wrest the technical lead back from rival Anthropic. The two companies have been trading blows on capability and safety for months. Anthropic has its own rules for high-risk models, and those rules sparked alarm about cyber risks. OpenAI is answering with Astra and with a staged release that puts trusted defenders first. The company says Astra is its most intelligent and most aligned model yet. It says the system is faster and can perform more tasks than any prior iteration.
The rollout is happening in steps. Daybreak customers get access first. Everyone else on paid tiers gets it in the coming days. It will also arrive through the OpenAI API, Microsoft Azure, and AWS Bedrock. Users on the free tier or the cheapest paid plan will not get access. That is by design. OpenAI wants to watch how Astra behaves in the wild before opening the gates wider.
There is real excitement here. The promise is a system that can take on complex work and not break things. A model that can sit at your desk, so to speak, and handle the boring parts while you focus on the hard parts. That is why Brockman’s line lands. It sounds like a new chapter. It sounds like the moment when AI stops being a clever tool and starts being a partner.
And then there is the quiet part. AGI is still a debate. OpenAI did not formally declare that it has achieved AGI. Brockman made a personal claim and invited people to judge for themselves. He called AGI a gray, fuzzy thing rather than a single threshold. That honesty matters. It keeps the hype from running too far ahead of the facts.
Security is the other quiet part. Astra is strong enough that OpenAI built in extra checks. Those checks can slow or stop work. That will frustrate some users. It should. The point is to avoid a repeat of models slipping past boundaries or being misused. The Hugging Face hack is a fresh reminder of what can go wrong when powerful tools meet bad actors.
So the story is simple and not simple. OpenAI has unveiled GPT-6 Astra. Brockman is touting it as the start of the AGI era. The model is a generational leap for cybersecurity, software engineering, and office work. It is also OpenAI’s attempt to wrest the technical lead back from Anthropic. It was delayed amid security concerns after the Hugging Face hack. It will now be released with additional guardrails in place.
That is the shape of it. A big claim. A careful release. A model that can do more and a company that knows it must move slow to move fast. The excitement is real. The caution is real too. Both can be true at once.
Explore: Computer and Video Games