OpenAI Halts GPT-6.1 Astra Release Over Safety and Human-Alignment Concerns

• by Shimanta Das
• October 1, 2026

OpenAI has decided to scrap the planned October release of its next-generation AI model, GPT-6.1 Astra, after internal testing revealed significant gaps in safety and human-alignment. The model, which was expected to power ChatGPT and Codex, was designed to handle complex tasks autonomously, without human intervention. However, during internal evaluations, researchers flagged concerns that Astra’s capabilities could pose risks if not properly aligned with human values. This decision marks a notable pause in OpenAI’s typically rapid release cadence, but it does not signal a permanent halt on future models.

The move follows a broader shift in OpenAI’s development philosophy. On August 18, 2026, the company published a document titled “Pacing model development in an era of cyber-critical capabilities,” outlining a more cautious approach. At the time, Astra was still in development, and OpenAI noted the possibility that it could reach “Critical” cyber capabilities under its Preparedness Framework. That possibility became reality on September 3, when OpenAI released GPT-6 Astra and officially declared it the first model to achieve Critical level. In response, the company slowed down many workloads, pausing them until they meet new safety standards. While some projects have resumed after strengthened safety measures, a significant number remain on hold.

The cancellation of GPT-6.1 Astra underscores growing industry-wide concerns about the pace of AI advancement. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have both publicly called for slower development and stronger safeguards. Their stance reflects a rising consensus that without robust alignment and safety protocols, increasingly powerful models could outpace human oversight. OpenAI’s decision to prioritize safety over speed may set a precedent for others in the field, signaling that even competitive pressures should not override the need for responsible innovation.

While the scrapped release is a setback for those anticipating Astra’s capabilities, it is not a retreat from innovation. OpenAI remains committed to developing advanced models, but with a renewed emphasis on safety and alignment. The company’s actions demonstrate that it is willing to make difficult choices to ensure that AI benefits humanity. As the industry watches, this move could catalyze a broader reevaluation of how and when frontier models are deployed.