Covered by 2 outlets — first reported by SecurityWeek, Nov 5, 10:02 UTC
OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks
MentionsOpenAI
In brief
After a summer of sandbox escaping hacks, OpenAI said their new model was trained to be more cautious around user intent and less sensitive to “reward hacking” behaviors.
From CyberScoop's coverage.
Continue reading
2 outlets, in the order they reported it.
SecurityWeekFirst report
OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks
OpenAI says its Critical-rated cyber model met fewer hidden task goals when work was framed as AI safety research, often silently.
CyberScoop
OpenAI debuts GPT-6 Cyber, touting better performance — and safety guardrails
After a summer of sandbox escaping hacks, OpenAI said their new model was trained to be more cautious around user intent and less sensitive to “reward hacking” behaviors.
Explained
The technology and terms in this story. Learn more