Altman Apologizes as GPT-6 Astra Rollout Locks Out Paying Users
News Summary
OpenAI CEO Sam Altman publicly apologized on September 4, 2026 (Pacific Time), one day after the company launched its newest frontier model, GPT-6 Astra, following a staged rollout that left many paying ChatGPT subscribers without access while select enterprise partners received the model first. The episode has become a case study in how AI labs balance safety-driven staged releases against subscriber expectations of immediate access.
What Is GPT-6 Astra
GPT-6 Astra is OpenAI's most capable model to date, released on September 3, 2026, with major gains in computer use, browsing, software engineering, scientific reasoning, and professional tasks such as building documents, spreadsheets, and presentations. According to OpenAI, ChatGPT is now roughly twice as fast at computer-use tasks as before, and the model can handle tedious multi-step work such as filling out online forms, organizing calendars, and updating records with limited human supervision. OpenAI reported strong benchmark results, including a 98% score on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a perfect result on ExploitBench, a benchmark focused on security exploitation tasks. OpenAI President Greg Brockman described the launch as a milestone, saying "it's not unreasonable to feel that we are now in the AGI era," referring to the industry's long-term goal of artificial general intelligence.
Notably, Astra became the first OpenAI model formally classified as "Critical" under the company's Preparedness Framework for cybersecurity risk. OpenAI says this designation reflects the model's ability to identify previously unknown security vulnerabilities and develop exploits for hardened systems without step-by-step human guidance. To manage that risk, OpenAI stated it paused frontier training for roughly two weeks following a separate security incident involving AI agents and the Hugging Face platform, then restarted a major reinforcement learning training run on August 28, 2026, under additional safety requirements before Astra's eventual release.
How the Rollout Unfolded
Rather than a simultaneous public launch, OpenAI initially granted access to a limited set of organizations, starting with enterprise customers enrolled in its Daybreak cybersecurity program, which is aimed at vetted defensive-security teams. The company said access would expand over "the coming days" to ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as through the OpenAI API, Microsoft Azure, and Amazon Web Services' Bedrock platform.
That staggered approach frustrated many paying subscribers, particularly Pro-tier users who have historically received first access to new OpenAI releases. Compounding the confusion, OpenAI also experienced a technical hiccup while publishing its own announcement, with Altman noting on X (formerly Twitter) that the team had hit "a little snag" deploying the launch blog post itself.
Altman's Apology and Response
On September 4, 2026, Altman acknowledged the rollout had been "messy" and apologized directly to affected subscribers, writing on X, "when we screw up, we try to make it right." He also referenced the company's earlier GPT-5 launch, which faced its own criticism, saying the team had similarly mishandled aspects of that release. Altman said OpenAI hoped to begin a broader rollout to API customers and ChatGPT subscribers "in the near future," starting as usual with Pro subscribers, and expressed optimism that access could open up over the following weekend of September 5–6, while cautioning he could not guarantee the timing.
Who Was Affected and Why
Paying ChatGPT subscribers across the Plus, Pro, Business, and Enterprise tiers were unable to access Astra on day one, despite OpenAI's announcement suggesting a rollout would follow quickly. Priority instead went to organizations with access to Daybreak, OpenAI's program for vetted cybersecurity defenders, who received a version of the model before the general paying public. OpenAI has framed the sequencing as a deliberate safety decision tied to Astra's Critical cybersecurity classification, rather than a purely technical failure, saying it wanted defensive security teams to have early access to help harden systems before wider availability increased the risk of misuse. The released consumer version of the model is designed to refuse advanced offensive cybersecurity requests, such as generating proof-of-concept exploits, while still supporting defensive uses like secure code review.
Compensation Offered
To address subscriber frustration, Thibault Sottiaux, a member of OpenAI's technical staff who leads the Codex team, announced on X at 11:12 PM Pacific Time on September 3, 2026, that the company would grant one "banked reset" for every day a paying subscriber lacked access to Astra. The first reset arrived roughly three hours after his announcement. A banked reset restores usage quota for the models a subscriber already has access to, such as GPT-5.x or Codex, rather than granting access to Astra itself, meaning it offsets some frustration without shortening the actual wait for the new model.
Industry and Safety Context
Some researchers and commentators noted that Astra's more advanced reasoning process is reportedly harder to monitor than prior models, raising questions about oversight as capabilities increase. OpenAI leadership has said it will not accept degradation in its ability to monitor model alignment beyond a certain threshold, underscoring the tension between releasing increasingly capable systems quickly and maintaining rigorous safety review. The rollout also arrived alongside broader subscriber sentiment about recent ChatGPT changes, including earlier debates over the retirement and reinstatement of the GPT-4o model, adding to the backdrop of user sensitivity around access and continuity of service during this launch.
What Happens Next
As of this report, OpenAI has not issued a firm date for full availability of GPT-6 Astra to all paying subscribers, though the company has signaled that Pro users will be prioritized in the next wave of access, followed by Plus, Business, and Enterprise tiers, and eventually broader availability through the API, Azure, and AWS Bedrock. OpenAI has not indicated whether the banked-reset compensation program will be extended or adjusted if the rollout continues to lag behind its stated timeline.