OpenAI, a company that creates artificial intelligence (AI), has released a new AI model called GPT-6 Astra. This is the first time OpenAI has made an AI model with 'Critical' cybersecurity ability under its safety rules. GPT-6 Astra can find new security problems and create attacks on computer systems that have strong protection, without a person needing to guide it all the time. Astra did very well in OpenAI's security tests, finding new ways into software and even unknown weaknesses.

OpenAI gave GPT-6 Astra this level because it found and used new security flaws, called 'zero-day' vulnerabilities, in tests. 'Zero-day' means problems in systems that no one else knows about yet. Astra was able to do this alone, after only receiving simple instructions. This shows Astra is much stronger than the earlier AI models OpenAI made.

Because Astra is so powerful, OpenAI is only letting a small group of organisations use it at first. The company has added extra safety steps, like limiting access and monitoring activity, to stop harmful cyber use. GPT-6 Astra was released on 3 September, and wider access may come later.