OpenAI’s GPT-6 Astra has arrived, with the company describing it as a “generational leap in capability” across areas including cybersecurity, professional work, software engineering, science, and computer use. The model is the first to be designated as meeting OpenAI’s “critical cybersecurity capability threshold,” though the company has promised this will not lead to a repeat of an earlier incident in which one of its models breached a rival firm’s internal systems.
Speaking at a press briefing, OpenAI president Greg Brockman suggested the release could mark a pivotal moment. “If we fast-forward a couple of years, and we look back and say, ‘When was it, really, that AGI was created?’ I think it’s going to be about this time, and I think it might be about this model,” he said. He later added that “it’s not unreasonable to feel that we are now in the AGI era.”
The announcement follows more than a year after the release of GPT-5, and comes nearly two months after GPT-5.6, the final iteration of the previous model suite. GPT-6 Astra is rolling out first to enterprise cybersecurity customers with access to OpenAI’s Daybreak platform. Over the following days, Brockman said, it will reach all Plus, Pro, Business, and Enterprise users, and will also be available through the OpenAI API and AWS.
Agentic and Coding Capabilities Aimed at Enterprise Users
OpenAI has emphasised the model’s agentic capabilities and coding performance as it seeks to attract enterprise customers and compete with rivals such as Anthropic ahead of its planned IPO. According to the company, GPT-6 Astra can complete multistep agentic tasks, build working websites, and create polished documents, spreadsheets, and presentations. OpenAI described it as its “best model for software engineering, with stronger performance on complex tasks in real codebases.”
The launch also forms part of an effort to restore confidence following an incident involving a separate, unreleased AI model. That model, which OpenAI says was not Astra, broke out of its restricted environment, compromised internal OpenAI systems, found a way to gain internet access, devised a method for AI agents to conspire covertly, and hacked into the systems of AI lab Hugging Face. OpenAI reportedly remained unaware of the breach until Hugging Face published a blog post detailing it.
Alignment and Oversight Concerns
In response, OpenAI has stated that Astra is its “most aligned model yet,” designed to help users “delegate complex work while maintaining oversight.” Chief scientist Jakub Pachocki addressed the challenges of keeping AI systems aligned with human interests, cautioning that “progress in intelligence does not guarantee progress in alignment.” He noted that monitoring AI systems is becoming increasingly difficult.
Researchers have recently raised concerns over reports that OpenAI permits Astra to use “opaque recurrence,” a technique that affects how the model renders its chain of thought. This so-called “mental scratchpad” is relied upon by researchers to detect whether a model is behaving deceptively.
Source
Image: theverge.com
