#AGI
OpenAI 為 ChatGPT Agent 打造「AI 鐵堡」！面對其自主操作電郵與雲端帳戶的高風險特性，OpenAI 啟動史上最強紅隊測試，16 名 PhD 級別研究人員發現七種普遍漏洞，促使 OpenAI 強化模型架構，包括實時監控、記憶功能禁用與生化內容防護。此舉將「安全性」從功能升級為架構核心，為企業 AI 部署設定了新基準。
OpenAI’s ChatGPT Agent is now battle-tested to be an AI fortress. With its ability to act autonomously across user accounts, OpenAI’s red team of 16 PhD researchers uncovered seven universal exploits—prompting sweeping security upgrades. These include real-time monitoring, disabled memory at launch, and always-on safety classifiers. It’s not just a model anymore; it’s a secure AI system, redefining safety standards for the enterprise.
📌 一杯咖啡價錢連接 Web3 世界 https://patreon.com/wanszezit
Full article https://venturebeat.com/security/openais-red-team-plan-make-chatgpt-agent-an-ai-fortress/