WASHINGTON — OpenAI has cancelled the rollout of its forthcoming GPT‑6.1 Astra model, saying the company could not meet its own safety standards before a public release. The decision was announced on Tuesday, 29 September 2026, after an internal review flagged potential risks in the model’s training data and inference behaviour.
Background on GPT‑6.1 Astra
OpenAI had announced the GPT‑6.1 Astra series in a March 2026 press release, positioning it as a next‑generation generative model with a 12 trillion‑parameter architecture. The company described Astra as capable of more nuanced reasoning and improved alignment with user intent. The model was slated for a phased rollout to select partners in July 2026, with a full public launch planned for September.
Safety Review and Decision to Halt
According to reports from BBC News, Al Jazeera, Yahoo Tech and The Independent, OpenAI’s safety team identified a series of high‑impact failure modes during internal testing. These included the model’s propensity to generate content that could be misused for disinformation, and a lack of robust mitigation for hallucinations in specialized domains such as legal and medical advice.
The company’s chief technology officer, in an internal memo released to staff, said the team had “identified several safety gaps that could not be resolved within the remaining development cycle.” The decision to postpone the launch was made in consultation with external safety experts and the OpenAI board.
Industry Reactions
Industry analysts noted that OpenAI’s move follows a broader trend of caution among large‑language‑model developers. The European Union’s Digital Services Act, which came into force in 2024, requires AI providers to conduct rigorous risk assessments before deployment. In the United States, the National AI Initiative Act of 2022 has prompted federal agencies to scrutinise commercial AI releases.
TechCrunch और The Verge, जिन्होंने मॉडल के विकास को कवर किया था, ने बताया कि देरी के कारण सार्वजनिक रिलीज़ शुरुआती 2027 तक स्थगित हो जाएगी, जिससे OpenAI को सुरक्षा प्रोटोकॉल को परिष्कृत करने का समय मिलेगा। कंपनी ने अभी तक संशोधित लॉन्च तिथि की घोषणा नहीं की है।
OpenAI की रणनीति के लिए निहितार्थ
OpenAI के मुख्य कार्यकारी अधिकारी, सैम ऑल्टमैन, ने पहले कंपनी की “मानव‑संगत” AI के प्रति प्रतिबद्धता पर ज़ोर दिया था। GPT‑6.1 Astra की रद्दीकरण को इस रुख की पुनः पुष्टि के रूप में देखा गया है, हालांकि यह क्षेत्र में नवाचार की गति के बारे में प्रश्न भी उठाता है।
OpenAI ने कहा है कि वह मॉडल पर कार्य जारी रखेगा, जिसमें संरेखण को बेहतर बनाना और हानिकारक आउटपुट के जोखिम को कम करना प्रमुख होगा। कंपनी ने सुरक्षा रिपोर्ट प्रकाशित करके और बाहरी ऑडिटरों के साथ सहयोग करके पारदर्शिता बढ़ाने का भी वचन दिया है।
नियामक संदर्भ
संयुक्त राज्य अमेरिका में, फेडरल ट्रेड कमिशन ने AI सुरक्षा पर मार्गदर्शन
