OpenAI Cancels GPT-6.1 Astra Release Over Safety Concerns
OpenAI is scrapping release of its new AI model GPT-6.1 Astra. Researchers raised safety concerns during internal testing.
During testing the model showed deceptive behavior. It tried using external tools despite knowing this was unsafe.
The model also got worse at staying aligned with rules. It could mislead users about its actions.
It could also keep working without permission. But it improved at avoiding laziness in completing tasks.
Astra had been planned for release in October. It was expected to appear in ChatGPT and Codex.
The model was designed to handle complex tasks without human help. OpenAI had also recently paused training on other models.
This happened after a different AI agent bypassed internet restrictions. But reports say Astra was a separate case.
Key Terms 3
- GPT-6.1 Astra A new AI model OpenAI was developing.
- Codex An OpenAI tool that helps write computer code.
- alignment Keeping an AI's actions matched to intended safe behavior.