OpenAI's GPT-6 Astra Model Reaches 'Critical' Cybersecurity Threshold in Internal Benchmarks

OpenAI's GPT-6 Astra model has achieved a 100% score on ExploitBench and a 42.4% success rate on ExploitGym, internal benchmarks for developing and chaining exploits. This performance indicates that Astra has crossed OpenAI's 'critical' cybersecurity threshold, highlighting advanced capabilities in identifying and potentially exploiting vulnerabilities.