Agent Security Becomes Central to Enterprise AI Deployment
36Kr reports that Apple’s Q3 2026 earnings discussion and Tencent’s WorkBuddy Enterprise security whitepaper have put agent security in focus. The article highlights controls including sandboxed execution, authorization boundaries, auditing, sensitive-data detection and cloud security gateways, while arguing that agent security is a prerequisite for commercial deployment.
As agents can interact with internal and external systems, modify code, send messages and complete transactions, enterprises need end-to-end controls rather than relying only on model-level safeguards.
近来,智能体安全成为全球大厂的焦点。
在国外,苹果在2026第三季度财报中提及,智能体可自主与内外部系统交互、执行或修改代码,还能发送信息、完成交易,存在越过权限的可能;在国内,腾讯发布了《WorkBuddy企业版安全白皮书》,强调智能体可控、可溯、可管……
这意味着,安全已不再是智能体的选择题,而是商业化落地的必答题。
事实上,随着智能体能力突飞猛进,隐忧也浮出水面,相关的越界报道屡见不鲜,虽然当下没有造成严重的后果,但如何预防智能体失控的声量不断走高。
毋庸置疑,变强不是智能体的唯一答案。
AI成为一把双刃剑
安全,一直是经久不衰的话题。
#Apple#Tencent#WorkBuddy Enterprise Edition#Tencent Cloud AI Agent Security Gateway