Unverified50% confidenceCautionExact time
In multi-agent systems, if all sub-agents share the same set of tools and access credentials, once a sub-agent is manipulated by a prompt injection attack, the attack surface extends to the entire system.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn multi-Agent scenarios, if each Agent uses its own idempotency key, the same business operation is treated by the system as multiple distinct requests.78% similarUnverifiedIn multi-agent systems, a compromised agent may pass a tainted context to other agents, creating trust-chain contamination, and the industry currently lacks a mature defense solution for this.77% similarUnverified攻击者通过提示词注入控制Agent的决策逻辑后，可利用Agent已获授权的所有工具和接口实现全链路攻击，形成混淆代理人（Confused Deputy）问题75% similarUnverifiedAssigning a unique identifier to each Agent resolves the risk that multiple Agents sharing the same set of credentials makes it impossible to identify the responsible party.74% similarUnverifiedMulti-Agent architectures require designing explicit message-passing interfaces and state synchronization mechanisms to ensure each Agent's output can be correctly consumed by downstream Agents74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/960437API
curl https://kongchang.com/api/v1/knowledge/claims/960437MCP
get_claim(id=960437)