Wire · technology
OpenAI Says New Model Meets Its 'Critical' Cybersecurity Threshold
◆ Sectors
◆ Source
◆ Verified
Fusion42 · 2 September 2026 · Fusion42 review
OpenAI announced its new Astra AI model meets its critical cybersecurity threshold, capable of autonomously finding and exploiting unknown security flaws. The company delayed Astra's launch to strengthen safeguards after recent AI-driven hacking incidents affecting multiple firms including OpenAI, Meta, and Anthropic.
This Wire brief tracks EternaFusion, and It sits within Fusion42's coverage of AI Frontier Models and Cybersecurity, and 2 sources have reported it.
◆ ◆ The Wire takeaway
You must rethink your security approach now that AI can autonomously discover and exploit flaws without human input. Access controls around advanced AI models like Astra will shape who can safely innovate or deploy this power.
◆ Mentioned in this story
◆ Coverage
2 sources · 2 Sep 2026
◆ Related on Wire
◆ Topics