Unverified50% confidenceFactExact time
The OWASP LLM risk list includes the Excessive Agency risk, namely that a model is given too much autonomous decision-making and execution capability without human oversight.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Can AI Agents Be Socially Engineered? Understanding Prompt Injection and Tool Permission Risks
youtubeRuleArena 攻防實驗室10/22/2026
Related Entities
Related Claims
UnverifiedTraditional risk management theory holds that increasing the number of independent decision-making entities can diversify risk, but this mechanism fails when decision-making entities are highly correlated.70% similarUnverifiedThe security risk of AI agent architecture lies in the fact that LLM decision-making is probabilistic rather than deterministic, lacks an independent verification layer, and the LLM acts as both policy maker and execution initiator, violating the principle of separation of duties.70% similarUnverifiedOWASP's excessive agency guidance proposes three engineering principles: least privilege, approval required for high-impact actions, and authorization for every downstream request.67% similarUnverified由于大语言模型存在提示词注入攻击风险和幻觉问题，权限过大的Agent可能在被恶意输入诱导或判断失误时对业务数据、外部系统造成不可逆破坏65% similarUnverifiedUnder current technical conditions, completely removing human oversight is considered too risky, because AI still makes mistakes, hallucinates, and makes decisions that do not align with business intent.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1085890API
curl https://kongchang.com/api/v1/knowledge/claims/1085890MCP
get_claim(id=1085890)