OpenAI is preparing to release Astra, its newest cyber-critical large language model. While the company plans to make the powerful AI model available soon, it will limit access to the software's most cutting-edge cybersecurity capabilities.
The decision to implement added security layers comes after internal testing found the model capable of executing complex cyberattacks with minimal human input. OpenAI is being extra cautious following the Hugging Face incident, even as it pitches its cybersecurity capabilities to potential customers.
As part of its rollout strategy, the company will give select partners early access to Astra. This measure is intended to give those partners sufficient time to shore up their defenses before the broader release.