OpenAI GPT-6 Astra Launched With Critical-Level Cybersecurity Capabilities as Part of AGI Push

OpenAI has released GPT-6 Astra, calling it the most capable flagship AI model yet. Its deployment is part of the San Francisco-based company's plans to eventually reach AGI (artificial general intelligence), a stage where a computer system can learn, reason, and perform any intellectual or cognitive task that a human can. GPT-6 Astra is designed for reasoning, coding, research and agentic workflows. It comes with a 1.05-million-token context window and supports up to 128,000 output tokens through the API.

GPT-6 Astra Brings Cybersecurity, Safety Upgrades

In a blog post, OpenAI said that GPT-6 Astra can find previously unknown security flaws and develop ways to exploit them across multiple well-protected systems. It is claimed to be built for the hardest end-to-end work, including complex reasoning, software engineering, computer use, research and document creation. The model supports reasoning effort levels from low through max, while its API supports text and image inputs, function calling, structured outputs and streaming.

GPT-6 Astra is claimed to be the first model to reach the Critical level under its Preparedness Framework. This essentially means that it can identify previously unknown vulnerabilities and develop exploit strategies across hardened systems with limited human guidance.

Citing an internal evaluation, the ChatGPT maker said that GPT-6 Astra achieved a 100 percent score on ExploitBench. Astra is also claimed to have built a browser-compromise chain that escaped a sandbox and executed commands on the host, while another test saw it combine vulnerabilities to create a local privilege-escalation chain.

Due ot its capabilities, OpenAI has introduced additional safeguards around the AI model. It is said to be significantly more resistant to jailbreaks than GPT-5.6 Sol, refusing 91.5 percent of requests in its cyber-jailbreak evaluations compared with 59 percent for the previous model. It is also said to have received roughly half as many higher-severity misalignment flags in a simulation involving more than 54,000 internal Codex tasks.

There are still limitations, though. As per the company, the GPT-6 Astra can can sometimes evade chain-of-thought monitoring in adversarial tests. These include instances when strategically underperforming or attempting certain sabotage tasks. The company says it has not found evidence of steganographic reasoning, but acknowledges that monitorability has decreased compared with GPT-5.6 Sol.

GPT-6 Astra has an April 30, 2026 knowledge cutoff. It is initially rolling out to enterprises through its Trusted Access Programme, with access through the API and Plus, Pro, Business and Enterprise plans expected in the following days.