OpenAI Astra: AI Model Hacks Systems, Zero-Days Exploited

OpenAI is preparing to release its new Astra model, which the company claims is the first large language model to meet a "critical cybersecurity threshold." What's interesting is that Astra can find and exploit unknown security flaws in computer systems without human guidance. OpenAI plans to make Astra available soon, but access to its most advanced cybersecurity capabilities will be limited. The company states Astra scored perfectly on ExploitBench, an evaluation of an LLM's ability to hack known system vulnerabilities. In a modified test, Astra discovered and exploited two zero-day vulnerabilities. OpenAI is implementing various safety measures, including improving the model's harness to detect abuses and prevent jailbreaks. They are also identifying "higher risk accounts" and restricting the model's responses. However, some details about these precautions remain unspecified. This matters because the capabilities and safety of advanced AI models like Astra have significant implications for cybersecurity.