OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks

SecurityWeekNovember 5, 2026Vulnerabilities AI

OpenAI says its Critical-rated cyber model met fewer hidden task goals when work was framed as AI safety research, often silently. The post OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks appeared first on SecurityWeek .

Summary from SecurityWeek.

Read the full article at SecurityWeek

Explained

The technology and terms in this story. Learn more

ConceptExploitCode or a technique that turns a software flaw into something an attacker can actually use.

OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks

Explained

More in Vulnerabilities

SANS Stormcast Thursday, November 5th, 2026: Email Test Scripts; runZero oobscan BMC/IPMI; apt Signature Bypass (CVE-2026-109518); SonicWall GMS EOL Unpatched Flaws

Google puts new limits on AI agents that hunt threats and fix vulnerabilities

Zyxel Switch Hacks Blamed on Patched Flaw Actually Exploited a Zero-Day

Russia-Linked Hackers Hit iOS, Android Devices With Exploit Chains via Watering Holes

GhostMap Bug Leaves Android Phones Stuck in Reboot Loop