OpenAI: Upcoming artificial intelligence model Astra will be the first with “critical” level cybersecurity capabilities

The OpenAI startup has been forced to consider the rapid development of its AI models and the extent of its impact on the cybersecurity field, so the company’s upcoming Astra has been given a capability rating of more than “critical” according to internal classifications. This means Astra is capable of finding new ways to cause significant damage to information infrastructure.

Image credit: Unsplash, Mariaia Berezovsky

According to OpenAI representatives, the Astra model is ready to find previously unknown types of vulnerabilities and exploit them without the need for step-by-step instructions from a human. This means that, judging from the internal classification of the readiness framework, the AI ​​model has the most advanced ability to invade the information systems of potential victims. Recognizing this fact does not negate OpenAI’s intention to present the model to the public in the near future, but rather to access its capabilities in the field of cybersecurity It would be more restrictive than she initially expected.

The previous complexity level (“High”) meant that AI models could exploit existing types of vulnerabilities and harden them to cause severe damage to information infrastructure objects. OpenAI explained in a post on the company’s blog page that it will release more details about the security of the Astra model when it is officially announced. Following the precedent of attacks on Hugging Face infrastructure, public attention to OpenAI’s activities in this area has increased significantly. By introducing the necessary restrictions, OpenAI expressed confidence that the risk of serious damage after the Astra model is released will be significantly reduced. Astra’s most advanced cyber security features will be available to members Dawn AllianceOpenAI noted.