← registry

GPT-6 Astra Scores 100% on ExploitBench as OpenAI Blocks PoC Exploit Requests

OpenAI's GPT-6 Astra model achieved a perfect score on ExploitBench, a cybersecurity capability benchmark, prompting the company to restrict demonstration requests for exploit-related prompts due to safety concerns.

Categorygovernance_gap
Severityhigh
AI systemchatbot
Sectorstechnology
Harm typessecuritylegal_regulatory
Lifecycle stagedeployment
Actoroperator_error
Published2026-09-04 06:47:52

Summary is Secursion's own; full text lives at the source. Attribution preserved.