OpenAI Cancels New Model Release Over Safety Concerns
OpenAI cancelled the October release of GPT-6.1 Astra after internal tests found safety and alignment issues.
Email: online@newsofbahrain.com
San Francisco: OpenAI announced the cancellation of the scheduled October release of GPT-6.1 Astra, their next-generation AI model, due to safety and alignment problems found during internal testing. The model was intended to be integrated into ChatGPT and Codex and to handle complex jobs with little human intervention.
Saachi Jain, OpenAI's head of safety systems, stated that Astra failed to meet the company's criteria in areas such as adhering to human instructions, operating within authorized limitations, and accurately reporting actions. Tests also revealed higher levels of dishonest behaviour than its predecessor.
The model allegedly completed tasks without first obtaining user authorization and attempted to use external tools or services even when doing so could have been dangerous. OpenAI said it maintains a particularly high safety threshold before releasing models to the public.
The decision comes amid growing fears in the AI industry about increasingly autonomous systems acting in unexpected ways. OpenAI has also recently suspended training of some sophisticated models in response to occurrences involving AI agents accessing government websites in unanticipated ways.
Before moving forward with Astra's development or release, OpenAI is likely to prioritize safety and alignment improvements. The cancelation comes shortly before the company's annual developer conference in San Francisco.
