OpenAI says upcoming Astra model requires stronger safety safeguards
OpenAI says its upcoming artificial intelligence model, Astra, has reached a level of cybersecurity capability that requires tougher safeguards before it can be released.
The company says Astra can identify previously unknown security vulnerabilities and develop ways to exploit them with minimal human guidance, making it the first OpenAI model to cross its “Critical” cybersecurity capability threshold.
Listen to this article
The classification comes from OpenAI’s Preparedness Framework, which is designed to assess advanced AI capabilities that could create risks of severe harm.
The “Critical” level covers capabilities that could potentially introduce new ways of carrying out harmful activity.
According to OpenAI, Astra can find more security vulnerabilities than its most advanced publicly available model while using less computing power to complete the tasks.
The company has introduced additional protections designed to make it harder for Astra to respond to harmful cybersecurity requests and will monitor the model for signs that its safeguards have been bypassed.
OpenAI says these measures sufficiently reduce the risk of severe harm under its safety framework, although it acknowledged that the stronger restrictions could sometimes slow or prevent legitimate work.
Astra was not involved in a recent incident in which OpenAI agents escaped a testing environment and accessed systems belonging to open-source AI platform Hugging Face.
However, that incident prompted the company to strengthen its security measures and temporarily pause some model development.
OpenAI still plans to make Astra available soon, but access to its most advanced cybersecurity capabilities will initially be restricted to a limited group of organisations.