← Back

Wire · opportunities

OpenAI rates Astra 'critical,' curbs hacking features before release

Published

2 September 2026

Topic

opportunities

Sectors

AI Frontier Models

Geography

United States

Source

Read at biz.chosun.com

Verified

Fusion42 · 2 September 2026 · Fusion42 review

OpenAI's AI model Astra achieved a 'critical' cybersecurity rating for its ability to autonomously discover and exploit zero-day vulnerabilities, prompting OpenAI to restrict its hacking features and implement stronger safety measures prior to public release.

This Wire brief sits within Fusion42's coverage of AI Frontier Models, and 4 sources have reported it between 1 Sep 2026 and 2 Sep 2026.

◆ The Wire takeaway

You must treat AI cybersecurity tools as potential threats alongside technologies to secure. OpenAI’s move forces founders building AI cyber tools to prepare for stricter access controls and compliance hurdles starting now.

Coverage

4 sources · first reported 1 Sep 2026 · latest 2 Sep 2026

Related on Wire

Topics

AI Frontier Modelsopenaiai-safetycybersecurityai-regulationastra