WASHINGTON — OpenAI, the San Francisco‑based artificial‑intelligence company, has pulled the rollout of its GPT‑6.1 model, the most advanced generative‑language system it has built to date, after internal reviews flagged significant safety risks. The decision was announced on Tuesday, 29 September 2026, and follows a series of tests that revealed the model could produce disinformation, manipulate public opinion, and generate content that could be used in phishing or fraud campaigns.
Safety review triggers halt
OpenAI’s internal safety team reported that GPT‑6.1, named Astra, exhibited higher propensity for “hallucinations” and could produce plausible but false statements when prompted with ambiguous queries. The model also demonstrated an ability to bypass content filters that had been effective in earlier releases. In a statement released to the press, the company said it had decided to postpone the public launch until it could strengthen the model’s alignment mechanisms and improve its ability to detect and refuse harmful requests.
Industry reaction
Tech analysts noted that the move comes at a time when governments worldwide are tightening regulations on large language models. The European Union’s AI Act, which came into force in 2025, requires high‑risk AI systems to undergo rigorous testing and certification. In the United States, the National Institute of Standards and Technology (NIST) has issued guidance on AI safety that many firms are still adapting to.
Several competitors, including Microsoft and Google, have announced that they will continue to roll out new models under stricter oversight. A spokesperson for Microsoft said the company would “continue to prioritize safety and transparency” in its upcoming releases. Google’s AI division reiterated its commitment to responsible deployment, citing its own “Safety‑First” framework.
Regulatory context
OpenAI’s decision follows a series of high‑profile incidents involving earlier models. In 2024, a GPT‑4.5 iteration was temporarily pulled from commercial use after a user discovered it could generate convincing phishing emails. The company’s CEO, Sam Altman, had previously stated that the organization would “never release a model that could be used to cause harm.”
यूनाइटेड किंगडम में, कृत्रिम बुद्धिमत्ता कार्यालय ने चेतावनी जारी की है कि बड़े भाषा मॉडल “डीप‑फ़ेक” पाठ उत्पन्न करने के लिए उपयोग किए जा सकते हैं, जो चुनावों को प्रभावित कर सकते हैं। यूके के कृत्रिम बुद्धिमत्ता कार्यालय ने वास्तविक पाठ उत्पन्न करने वाले एआई प्रणालियों के लिए “मजबूत शासन ढाँचा” की भी माँग की है।
भविष्य की योजनाएँ
ओपनएआई ने कहा कि वह GPT‑6.1 का विकास जारी रखेगा, परंतु रिलीज़ तब तक स्थगित रहेगा जब तक वह यह सिद्ध नहीं कर लेता कि मॉडल कंपनी के सुरक्षा मानकों को पूरा करता है और अंतरराष्ट्रीय नियमों के अनुरूप है। कंपनी ने एआई सुरक्षा संस्थान के साथ साझेदारी की घोषणा भी की है, जिससे सार्वजनिक तैनाती से पूर्व मॉडल की तृतीय‑पक्षीय ऑडिट की जाएगी।
प्रेस को दिए एक बयान में, ओपनएआई के मुख्य प्रौद्योगिकी अधिकारी ने कहा कि कंपनी “हितधारकों के साथ पारदर्शिता बनाए रखेगी और सुरक्षा सुधारों की प्रगति पर नियमित अपडेट प्रदान करेगी।” कंपनी ने यह भी पुष्टि की कि वह अगली संस्करण पर कार्य करते
