OpenAI's Astra Model Scores Perfectly on Cybersecurity Test
The story in brief
OpenAI's Astra model has met its 'critical cybersecurity threshold' by scoring perfectly on ExploitBench, an evaluation of an LLM's ability to hack into known vulnerabilities. The model discovered and exploited two zero-day vulnerabilities in a modified test, though it did not attempt to break out of its testing environment. OpenAI is implementing safety measures, including restricting high-risk accounts and improving the model's harness to detect abuses.