Unverified50% confidenceCautionExact time
The main engineering risks of Agents include goal drift, tool misuse leading to infinite loops, unauthorized execution, and privacy leakage caused by placing sensitive data in prompts.
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter agents gain tool-calling and external system access capabilities, the attack surface expands significantly, facing risks such as prompt injection, unauthorized operations, and data leakage76% similarUnverified在具备工具调用能力的 Agentic 系统中，一次成功的提示注入可能导致数据泄露、恶意代码执行甚至供应链攻击75% similarUnverified使用模型代理工具存在数据隐私风险（代码经第三方服务器转发）和延迟影响（引入数十毫秒网络延迟）73% similarUnverifiedThe Ministry of Industry and Information Technology issued a security warning regarding a type of AI Agent tool, listing multiple vulnerabilities.73% similarUnverified使用他人打包的Skill文件存在安全风险，可能引入恶意指令73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931874API
curl https://kongchang.com/api/v1/knowledge/claims/931874MCP
get_claim(id=931874)