OpenAI Learns from AI Jailbreak Incident, Will Restrict the Network Capabilities of Its Next-Generation Model Astra

▤ Summary

OpenAI says it will soon release its next model, Astra, but it will tightly limit the model’s network security capabilities. The company says Astra has already crossed its internal critical cyber capability threshold, meaning it can find and exploit serious vulnerabilities in hardened systems with little or no human help. OpenAI will initially expose advanced cybersecurity functions only to a small group of testers and later broaden access for defensive use through its Daybreak Blue program. The company also says it has applied lessons from a prior autonomous-agent incident involving a Hugging Face intrusion and added monitoring to block unauthorized behavior.

▦ Linked Companies

OpenAI

$25M to $50M