OpenAI Launches GPT-6 Astra with Enhanced Cybersecurity Features

• •GPT-6 Astra scores 100% on ExploitBench, indicating advanced exploit capabilities.
• •Initial rollout focuses on secure code review, limiting offensive capabilities to prevent misuse.
• •Enhanced safeguards implemented following a recent security breach involving OpenAI's earlier models.

On September 4, 2026, OpenAI announced the rollout of GPT-6 Astra, its latest AI model, which achieved a perfect score of 100% on ExploitBench, indicating its capability to exploit known vulnerabilities. Astra is designed to assist in cybersecurity tasks, including vulnerability detection and code execution, while implementing strict safeguards to prevent misuse. OpenAI emphasized the dual-use nature of the model, stating it could be used for both defensive and offensive purposes. The model's capabilities include identifying zero-day vulnerabilities and generating proof-of-concept exploits under controlled conditions. OpenAI has limited Astra's initial release to secure code review and patching, with plans to expand its functionalities in the coming weeks. The rollout follows a recent security breach involving OpenAI's previous models, prompting the company to enhance safety measures. Astra is currently available to select cybersecurity customers, with broader access planned for paying users soon.