OpenAI released GPT-6 Astra on September 3, describing it as the company’s first model to reach the Critical level of cybersecurity capability under its Preparedness Framework.

The company said that, with the right tools and access, Astra can identify previously unknown flaws and develop new ways to exploit many well-protected systems without a person guiding each step.

OpenAI said it strengthened protections against harmful cyber actions and added stricter isolation, checkpoint encryption, full-trajectory monitoring, including chain of thought, and broader misalignment monitoring for external tool-using inference.

The company also said internal tests found Astra was harder to jailbreak and received about half as many higher-severity misalignment flags as GPT-5.6 Sol in a simulation using more than 54,000 internal Codex tasks.

OpenAI also said Astra is harder to monitor than Sol because it is better at controlling its chain of thought and can sometimes evade internal monitors in adversarial tests.

The release comes as AI safeguards remain under scrutiny after leading tech firms agreed to White House AI safeguards in 2023.

Those voluntary commitments later faced questions over how fully companies had implemented them.

🔗 Source: OpenAI

Recent OpenAI developments