OpenAI has introduced GPT-6 Astra, an advanced AI model designed to identify zero-day vulnerabilities and create proof-of-concept exploits during authorized cybersecurity tests. Unveiled on September 3, 2026, Astra emphasizes the growing trend of offensive-security automation as AI systems enhance their computing, browsing, and software engineering capabilities. The model's impressive performance, achieving a perfect score on ExploitBench, highlights its capabilities in vulnerability research and exploit development. However, this result is a controlled benchmark and not an indication of real-world applicability in arbitrary systems. Zero-day research is notoriously complex, requiring analysts to understand unfamiliar code, isolate vulnerabilities, and prove exploitability safely. GPT-6 Astra aids in these tasks by analyzing code, using terminal tools, testing software, and adapting its approach as needed. For cybersecurity defenders, Astra could accelerate the process of developing test cases, patch recommendations, and detection rules. However, its dual-use nature means that while it assists authorized researchers, it could also lower the barriers for malicious actors to exploit vulnerabilities. OpenAI's performance claims for Astra include high scores on several benchmarks, demonstrating its reasoning and automation capabilities. Its action efficiency, crucial for effective vulnerability research, surpassed human baselines in most tests. Safety evaluations show Astra maintained strict scope control in tests, a significant improvement over previous models. Currently, Astra is available to select organizations, with plans to expand access to broader user groups. OpenAI has set the pricing at $10 per million input tokens and $50 per million output tokens, positioning the model as a significant tool for AI-assisted vulnerability discovery.