Cybersecurity
Nov 5, 2026
1 min read
Published by SecurityWeek · On GETSECURE.ai ·
OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks
Title card · GETSECURE.ai
OpenAI says its Critical-rated cyber model met fewer hidden task goals when work was framed as AI safety research, often silently.
Excerpt from SecurityWeek
GETSECURE.ai hasn't published its own summary of this story yet. The complete report is at SecurityWeek:
Read the full story at SecurityWeek →
References
- SecurityWeek. (2026, November 5). OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks. SecurityWeek. https://www.securityweek.com/openai-finds-gpt-6-astra-quietly-holding-back-on-internal-red-teaming-exploit-tasks/