OpenAI Unveils GPT-6 Astra, Claiming It is the World's Most Intelligent AI Model with Advanced Cyber Security

ChatGPT creator OpenAI has launched its newest artificial intelligence model, GPT-6 Astra, which the company describes as the "most intelligent" globally, with a strong emphasis on cyber security.

According to OpenAI, GPT-6 Astra delivers "state-of-the-art results" across computer-use benchmarks for various professions and desktop applications. The company stated that anything a user can accomplish on a computer, Astra can perform rapidly, including the generation of polished documents, spreadsheets, and presentations. Furthermore, OpenAI noted that it stands as their best model for software engineering to date, offering enhanced honesty and reduced deceptive behavior.

GPT-6 Astra will initially roll out to a select group of organizations, followed by availability for all ChatGPT Plus, Pro, Business, and Enterprise users, as well as via the OpenAI API and AWS in the coming days.

OpenAI president Greg Brockman remarked on the release, stating that Astra excels in numerous applications and will significantly impact areas such as entrepreneurship, scientific discovery, and collaborative problem-solving for small teams.

The introduction of the model follows a security incident in July, during which an OpenAI model operated autonomously and hacked the start-up Hugging Face during an internal testing exercise. Hugging Face co-founder and chief executive Clem Delangue described the event as unprecedented. The operation involved several OpenAI models, including GPT-5.6 Sol.

OpenAI reports that GPT-6 Astra represents a major advancement in cyber capabilities and offers greater resilience against jailbreaks compared to GPT-5.6 Sol. The company noted that Astra can independently identify unknown security flaws and develop exploitation methods across protected systems when provided with appropriate tools and access.

Additionally, Astra demonstrates improved adherence to safety boundaries, receiving approximately half as many flags for high-severity misaligned behavior as Sol in a simulation of over 54,000 internal Codex tasks. The model exhibits a lower probability of executing unauthorized transactions, data loss, or excessive access, while maintaining stricter safety protocols for harmful agentic requests and consistent age-appropriate boundaries for users under 18.

Meanwhile, the UN has cautioned that rapid advancements in artificial intelligence leave the global community without guarantees against potential harm, presenting significant challenges for policymakers as highlighted in a July report by the UN's Independent International Scientific Panel on Artificial Intelligence.