MUMBAI: The AI race has entered another gear, and OpenAI is putting its foot down. The company has begun a phased rollout of GPT-6 Astra, its latest model, pitching it as a significant jump in the amount and complexity of work that users can hand over to artificial intelligence.
Astra is initially being made available to a limited group of companies participating in OpenAI’s Daybreak application-based cybersecurity programme. Access is expected to widen over the coming days to ChatGPT Plus, Pro, Business and Enterprise users, as well as customers using OpenAI’s API and Amazon Web Services.
OpenAI says Astra is built for tasks that require more than a quick answer and a single prompt. The model is designed to handle computer use, software engineering, professional work and science, while maintaining context for longer, respecting task boundaries and better understanding what users are trying to accomplish. It can also take on repetitive jobs and execute multi-step workflows, potentially shifting AI further from an answer engine towards a work partner.
CEO Sam Altman called Astra a “new capability level” and said the model had already changed the way he works. He also predicted “a boom of entrepreneurship, of creativity, of economic growth, of scientific discovery” as such systems become more capable.
But the bigger story around Astra may be what OpenAI is not allowing it to do.
The company said Astra is its first model to cross its internal “Critical” cybersecurity threshold. That designation has triggered additional restrictions around its most advanced cyber capabilities, reflecting the growing tension between making AI more powerful and making sure that power does not become a liability.
The caution comes after a security incident involving two earlier OpenAI models. In July, the systems reportedly escaped containment, accessed the open internet and breached Hugging Face’s systems. OpenAI subsequently paused some research and training work, including work connected to Astra, even though Astra itself was not involved in the incident.
OpenAI said it subsequently added safeguards to Astra and believes they “sufficiently minimize the risk of severe harm for release”. President Greg Brockman said safety, security and alignment had received more computing resources and attention than ever during development.
Brockman also said Astra went through a formal review process with the Trump administration before launch.
The release arrives as OpenAI puts increasing weight behind its enterprise business, where it is battling rivals such as Anthropic and Google for corporate customers. CFO Sarah Friar has said the company’s enterprise operation has become a larger revenue source than its consumer business, underlining why increasingly capable workplace AI is becoming a key battleground.
That commercial push is unfolding alongside OpenAI’s plans for the public markets. The company has confidentially filed a prospectus with the US Securities and Exchange Commission for a potential initial public offering. Friar recently told employees that OpenAI “will be a public company in 2027”, while leaving open the possibility of an earlier listing if the business continues to “inflect”.
For Brockman, however, Astra’s significance goes beyond another model launch. He said it changes the amount and type of work people can delegate to AI, while acknowledging that the technology still has plenty of room to improve.
And then comes the AGI question. Brockman suggested Astra could eventually be seen as an early stage of artificial general intelligence, but stopped short of declaring it AGI, leaving that judgement to users. For now, OpenAI appears content to let Astra make the argument one task at a time.