Unverified50% confidenceSolutionExact time
Mohi Uddin described an attack path called protocol pivoting: an attacker plants malicious text in tool output, the first agent passes it to a second agent as a normal task, and the second agent executes it because it trusts the sender.
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMaliciously crafted tool return values can induce an agent to perform unintended operations, i.e., prompt injection attacks68% similarUnverifiedMisalignment behaviors in tool-calling agents amplify the leverage of risk because they can trigger external actions such as sending emails, modifying databases, and calling APIs.67% similarUnverified代理系统在长对话链中反复调用工具时的行为类似于部分可观测马尔可夫决策过程（POMDP），存在目标漂移现象67% similarUnverified这类系统多采用人在回路或人在回路之上的设计，由人类操作员做出攻击决策并选定目标，AI仅负责末段追踪与撞击65% similarVerifiedJailbreak techniques can be categorized into prompt injection, adversarial prefix attacks, multi-step context construction, and automated jailbreaking.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/988945API
curl https://kongchang.com/api/v1/knowledge/claims/988945MCP
get_claim(id=988945)