OpenAI has postponed the release of GPT-6.1 Astra after the model underperformed on safety evaluations compared to its current version. The decision marks a cautious approach to deploying the next-generation AI system.
OpenAI is holding back a version of its Astra artificial intelligence model due to safety evaluation shortcomings. The unreleased GPT-6.1 Astra failed to meet the safety standards of the company's existing model, prompting the delay.
Despite improvements in other areas, the new version did not clear the safety benchmark required for public release. The model showed progress in addressing so-called model laziness—instances where AI systems fail to complete assigned tasks or provide incomplete responses.
This move reflects OpenAI's stated commitment to safety testing before deploying new AI systems. The decision comes as major AI developers face increasing scrutiny over the safety and reliability of their models.
The timing of a revised release remains unclear. OpenAI has not announced specific plans to resubmit an improved version for evaluation.
The Astra model represents OpenAI's latest generation of artificial intelligence technology. The company had been expected to debut GPT-6.1 Astra as the next iteration in its flagship GPT series, which powers ChatGPT and other applications.
The delay signals that performance gains alone are insufficient for OpenAI's release criteria. Safety evaluations now serve as a critical gating factor in the company's development timeline.
Industry observers note that such delays have become more common as AI capabilities advance and regulatory pressure increases. Companies are investing more resources in safety testing to avoid potential issues post-release.
OpenAI's decision aligns with its public position on responsible AI development. The company has previously emphasized the importance of comprehensive testing before deploying powerful AI models to users.
Florida's Attorney General is asking a judge to block OpenAI from giving ChatGPT human attributes, arguing the AI's use of first-person language and emotional mimicry deceive users into false trust.
Manus has launched version 2.0 of its AI agent platform, enabling users to edit videos, host multiplayer games, and deploy personal agents directly from their phones.
Harvard psychologist Steven Pinker argues that AI safety requires pragmatic engineering rather than doomsday predictions. He declined a public debate on existential risks, viewing such events as unproductive spectacle.
As artificial intelligence increasingly tackles complex mathematical problems through brute computational force, mathematicians are grappling with how to preserve the creative and artistic dimensions that have defined their field for centuries.