Verified65% 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.
3
Sources
65%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCommon security risks of self-hosted AI agents include prompt injection, improper key management and missing auditing.77% similarUnverifiedDifferent types of tools have different risk surfaces: the main risk of retrieval tools is privacy leakage, action tools produce irreversible side effects, compute tools carry the risk of resource exhaustion, and write tools carry the risk of data corruption or loss.77% similarUnverifiedAfter 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% similarUnverifiedFor tools like RIA, supply chain risks include vulnerabilities in the tool itself or the tool being tampered with, as well as data leakage caused by analyzed code snippets possibly being uploaded when the AI Agent goes online.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931874API
curl https://kongchang.com/api/v1/knowledge/claims/931874MCP
get_claim(id=931874)