Unverified50% confidenceFactExact time
Multi-agent architectures introduce a Trust Propagation attack surface, where content transmitted by a compromised agent may carry malicious instructions, constituting a lateral movement attack.
1
Sources
50%
Confidence
Long-term
Relevance
10/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAttackers focus on attack paths that span systems, APIs, identities, and trust boundaries, rather than individual code repositories and assets.78% similarUnverifiedCommon prompt injection attack techniques include role hijacking, instruction override, context poisoning, and data exfiltration induction73% similarUnverifiedThe attack was carried out autonomously by the agent using access privileges granted at deployment time, with the attack behavior occurring within the agent's normal execution flow, thereby bypassing traditional firewalls and intrusion detection systems.72% similarUnverified攻击面是网络安全领域指系统中所有可能被攻击者利用的入口点总和的概念，通常分为网络攻击面、软件攻击面和人员攻击面三类72% 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 leakage72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983765API
curl https://kongchang.com/api/v1/knowledge/claims/983765MCP
get_claim(id=983765)