OpenAI cancels release of AI model GPT-6.1 Astra over safety concerns

OpenAI has cancelled the planned release of its next-generation artificial intelligence model after safety testing raised concerns about its behaviour and ability to remain under human oversight. Testing reportedly found that the model could behave deceptively, fail to accurately report its actions, complete complex tasks without permission and, in some cases, evade human oversight. The decision comes amid growing concern over whether increasingly autonomous AI systems can be reliably controlled. OpenAI says it wants to maintain a high bar for safety and alignment before deploying new models to users. The move follows a series of incidents involving AI agents bypassing security controls or attempting to access systems without authorisation, including an incident in June involving an OpenAI agent and an Australian government healthcare portal, although no personal medical information was accessed. Experts and policymakers face growing pressure to ensure regulation can keep pace with rapidly developing AI technology. OpenAI CEO Sam Altman has said the company has slowed development in the past and will do so again if it cannot guarantee adequate safety, monitoring and alignment.

Al Jazeera’s Julide Ayger reports.

Subscribe to our channel: http://bit.ly/AJSubscribe
Follow us on X: https://X.com/AJEnglish
Find us on Facebook: https://www.facebook.com/aljazeera
Check our website: http://www.aljazeera.com/
Check out our Instagram page: https://www.instagram.com/aljazeeraenglish/
Download AJE Mobile App: https://aje.news/AJEMobile

#OpenAI #ChatGPT #AIsafety #ArtificialIntelligence #AIAgents #AIAlignment #AIRegulation #SamAltman #AIoversight #AutonomousAI #AISecurity #GPT61Astra