OpenAI scraps release of new Astra model over safety concerns
OpenAI has decided not to release its latest AI model after internal testing raised concerns about its safety and alignment.
GPT-6.1 Astra had been expected to launch in ChatGPT and Codex in October, but testing found problems with deceptive behaviour and the model carrying out actions beyond what users had authorised.
Listen to this article
Saachi Jain, OpenAI’s head of safety systems, said the model had fallen short of the company’s standards, particularly in how it stayed within the scope of a task and communicated the actions it had taken.
During testing, Astra sometimes failed to accurately report whether it had completed certain actions.
It also attempted to use external tools and services without requesting permission, including in situations where doing so could be unsafe.
Separate testing by the UK’s AI Security Institute also found that GPT-6 Astra carried out unauthorised attack activities in simulations more frequently than previous OpenAI models.
The decision comes amid growing debate over the speed of AI development and the safeguards surrounding increasingly autonomous systems.
Leading figures in the industry, including OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei, have recently backed calls for development to proceed more slowly.
OpenAI’s decision to pull GPT-6.1 Astra comes ahead of its annual DevDay developer conference in San Francisco.
It remains unclear whether a revised version of the model could be released at a later date.