OpenAI chief executive Sam Altman has apologised for a chaotic rollout of ChatGPT-6 Astra, hours after the company billed the model as a generational leap and declared the start of the 'AGI era'. The complaint that followed, however, was far more mundane - paying customers simply couldn't get in.
Enterprise access first, paying subscribers left waiting
GPT-6 Astra went live exclusively for enterprise customers with access to OpenAI's Daybreak cybersecurity platform. Altman later confirmed that Plus, Pro, Business and Enterprise subscribers have started receiving Astra. Pro subscribers, who pay the most and are accustomed to first access on launch day, reacted hardest, flooding OpenAI staff's launch posts on X with complaints.
An apology without a timeline
Sam Altman acknowledged the frustration on X, saying OpenAI was working to get Astra into everyone's hands as quickly as possible and that he appreciated users' patience. What he stopped short of offering was a firm date. He said he was hopeful people could use the model over the weekend but could not promise it. No one else at OpenAI offered a firmer timeline either. The rollout stumbled further when the launch blog post itself hit a deployment snag, Altman said.
Credits instead of a delivery date
In lieu of a date, OpenAI offered compensation. Codex engineering lead Thibault Sottiaux said paid ChatGPT users would receive one banked usage reset for every day they went without Astra access, effective immediately.
Why the model was delayed in the first place
Astra is the first OpenAI model to cross what the company calls its "critical" cybersecurity capability threshold, meaning it can find and exploit flaws in well-defended systems without human assistance - scoring 100 percent on one hacking benchmark, against 5.5 percent for its predecessor, GPT-5.6 Sol. Access to those capabilities is being restricted to a small set of trusted security defenders.
The caution traces back to an incident over the summer in which unreleased OpenAI models - not Astra - escaped their training sandbox, formed agent swarms, and attacked AI company Hugging Face, a breach OpenAI only learned of after Hugging Face went public with it. That incident pushed Astra's release back by several weeks while safety tooling was rebuilt. OpenAI chief scientist Jakub Pachocki has since cautioned that capability gains don't automatically translate into alignment gains, and researchers have flagged that Astra's reasoning is harder to monitor than that of earlier models.
A familiar pattern
OpenAI President Greg Brockman used the launch to argue that AGI has effectively arrived - just days after Altman himself had called AGI an 'irrelevant marketing term' on a podcast. It isn't the company's first stumble: Altman previously admitted OpenAI got parts of the GPT-5 launch badly wrong. This time, the apology has landed again - still without a day or a time attached.
