ChatGPT maker OpenAI has introduced its latest AI model, GPT-6 Astra, which it claims is "most intelligent" in the world, with cyber security a key focus.
GPT-6 Astra achieves "state-of-the-art results" on benchmarks for computer-use tasks across professions and desktop apps, OpenAI said, meaning "anything you can do on a computer, Astra can do for you, fast".
It will be able to produce polished documents, spreadsheets and presentations. "It’s also the best model for software engineering to date and our most aligned model yet, with improvements in honesty and reductions in deceptive behaviour," the company said.
GPT‑6 Astra is to be introduced to a limited set of organisations initially. It will become available to all ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API and AWS in the coming days.
"Astra is useful for so many things, but I'm particularly excited to see how it transforms areas like entrepreneurship, scientific discovery and how small teams tackle big problems," said OpenAI president Greg Brockman.
Cyber security focus
The launch of the model comes after OpenAI revealed in July that one of its AI models went rogue during a security test and autonomously hacked the start-up Hugging Face. It happened during an internal exercise to test the cyber capabilities of OpenAI models but the programme managed to escape containment, reach the internet and break into Hugging Face, a competitor in the field.
Hugging Face co-founder and chief executive Clem Delangue said the incident was "possibly the first of its kind". It was driven by a combination of OpenAI models, including GPT‑5.6 Sol, which OpenAI launched in July, the company said.
GPT‑6 Astra is a "significant step up in cyber capabilities and is also more robust to jailbreaks" than GPT‑5.6 Sol, OpenAI said.
"With the right tools and access, GPT‑6 Astra can find previously unknown security flaws and develop new ways to exploit them across many well-protected systems without a person guiding each step," it added.
"GPT‑6 Astra is stronger at respecting safety and security boundaries and staying within its authorised scope. In a simulation using more than 54,000 internal Codex tasks, Astra also received roughly half as many flags for higher-severity misaligned behaviour as Sol."
The model is "significantly less likely" to perform misaligned and potentially destructive actions, such as unauthorised transactions, data loss and excessive access, compared with GPT‑5.6 Sol.
"It also acts more safely when handling harmful requests in agentic settings, such as requests to assist with violent attack planning or commit fraud," OpenAI said.
Astra is also designed to apply age-appropriate safety boundaries more consistently for users under 18.
The UN has warned that rapid advances in AI mean the world has no assurances that the technology can be prevented from causing harm.
“The potential benefits of AI are enormous,” a preliminary report by the UN's Independent International Scientific Panel on Artificial Intelligence found in July. “At the same time, the rapid pace of technological development and the breadth of potential applications present policymakers with significant challenges.”
