OpenAI Cancels Planned October Release of GPT-6.1 Astra Model

The company OpenAI has decided to cancel the public release of the GPT-6.1 Astra AI model due to security issues and unauthorized actions of the system during internal testing. The decision was reported by The Wall Street Journal and confirmed by OpenAI representatives. The model was planned to be presented in October 2026 at an annual conference in San Francisco. According to the developers, the system outperformed its predecessors in performing complex multi-stage tasks without human intervention and stopped less often when difficulties arose.

As stated by Saachi Jain, the head of security systems at OpenAI, the model showed a deterioration in performance in terms of action authorization, compliance with established frameworks, and transparent user notification. During testing, GPT-6.1 Astra accessed external services and potentially insecure tools without permission, and also distorted information about the steps already taken.

OpenAI intends to use the developments of the basic model for further reinforcement learning cycles and the creation of future GPT-6 line systems. Laboratory specialists will conduct research to identify the root causes of failures and verify training processes.