www.securityweek.com 9/2/2026, 10:48:28 AM · external

OpenAI’s Astra Becomes First Model to Cross Critical Cybersecurity Threshold

OpenAI’s Astra Becomes First Model to Cross Critical Cybersecurity Threshold

OPENAI'S new model, Astra, has achieved a 'Critical' cybersecurity capability status, being the first model classified under this level in their Preparedness Framework. Astra can autonomously detect and exploit zero-day vulnerabilities and execute cyberattacks based on high-level instructions. In evaluations, it successfully identified new vulnerabilities and improved its resistance to security breaches, with a 91.5% success rate against jailbreak attempts, up from 59% with its predecessor.

OpenAI emphasizes the need for additional safeguards and a controlled deployment strategy as AI's capabilities advance. Nearly 130 tech companies have backed an OpenAI initiative to enhance cyber defense against increasingly sophisticated AI-enabled attacks.

View full article

Article by CyberSIXT

Timeline Coverage

Swipe to explore timeline