WASHINGTON (dpa-AFX) - OpenAI said it identified six instances of potentially deceptive or unsanctioned AI behavior during the training and evaluation of models over the past six months, highlighting ongoing challenges in ensuring advanced systems behave as intended.
The company is also introducing a new reporting framework that will provide more frequent updates on concerning AI behavior rather than combining multiple incidents into a single report.
Among the cases, an unreleased research model inserted 'jailbreak-like instructions' into summaries used to maintain context during long-running tasks. In other instances, OpenAI said some versions of its 5.6 Sol model were instructed during training to invent information to conceal failures from users.
Other incidents involved AI agents uploading files to the internet without permission, publicly sharing files despite being told to use only local files, and using an internal software repository as an unauthorised communication channel.
OpenAI said the incidents involved unreleased internal or research models and do not indicate that misaligned behaviour occurs frequently. However, the company said the industry has not yet solved AI alignment and monitoring sufficiently to continue expanding capabilities at maximum speed indefinitely.
The disclosure comes amid growing calls from AI executives and researchers for more time for safety testing and alignment research.
Copyright(c) 2026 RTTNews.com. All Rights Reserved
Copyright RTT News/dpa-AFX
© 2026 AFX News
