Sam Altman apologises for GPT‑6 Astra rollout delays – what went wrong and how OpenAI is responding

When OpenAI unveiled GPT‑6 Astra last month, the hype was palpable. The model promised a leap in AI‑driven cybersecurity, and the company bragged that it would soon be in the hands of every ChatGPT Plus, Pro, Business and Enterprise user. Reality, however, turned out to be a bit messier.

Instead of an instant rollout, access was first granted only to a handful of enterprise customers who already run OpenAI’s Daybreak security platform. The rest of the paying crowd – the folks on the various subscription tiers and developers using the API, Azure or Amazon Bedrock – were told they’d get in “over the coming days.”

That promise stretched into weeks, and social‑media chatter grew increasingly impatient. Users posted screenshots of empty dashboards, shared screenshots of error messages and, frankly, expressed a lot of frustration. The backlash eventually forced Sam Altman to step onto X (formerly Twitter) and own up to the misstep.

Altman’s note was brief but sincere. He admitted that the rollout “has not gone as planned,” offered an apology and said OpenAI was working around the clock to get Astra into the hands of all subscribers. He hinted that many might see access “over the weekend,” yet he stopped short of pinning down an exact date. In other words: a promise with a soft‑footed deadline.

To soothe the ruffled feathers, OpenAI announced a goodwill gesture. According to Thibault Sottiaux, who heads engineering for Codex, any paid ChatGPT user who remains locked out of Astra will receive a “banked usage reset” for each day of downtime. In plain English: if you’re paying and still can’t use the new model, you’ll get extra usage credit to make up for the loss.

The company also confessed that its own launch blog suffered a technical hiccup, adding another layer of embarrassment to an already shaky debut.

Why such caution this time? GPT‑6 Astra isn’t just a bigger language model; it’s been tuned to sniff out and even exploit vulnerabilities in highly secured systems. That level of power raises the stakes for safety and misuse, so OpenAI chose to limit early access to a trusted, smaller group while it tightens safety nets and runs additional evaluations.

In short, the excitement surrounding Astra is still very much alive, but the focus has shifted from what the model can do to when ordinary paying users will finally get to try it. For now, they have an apology, a promise of compensation and, unfortunately, an open‑ended wait.