Unverified50% confidenceFactExact time
An attacker can embed a prompt injection payload in the description field of an MCP tool, causing the model to execute it as a system-level instruction during inference; this is an observed real-world attack technique, not a purely theoretical exercise.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
Verified提示注入是一种针对大语言模型的攻击手法，攻击者通过在模型输入数据中嵌入恶意指令劫持模型的正常行为76% similarUnverifiedThe ability to make a model 'ignore all previous instructions' shares the same origin as the prompt injection problem in the security field; attackers exploit the model's tendency to obey instructions that appear later by embedding malicious instructions in web page content or documents to hijack the model's behavior.74% similarUnverifiedAdversarial distillation attackers systematically send large volumes of requests to the target model, collecting input-output pairs—especially responses containing complete reasoning chains—to train their own models and replicate the target model's capabilities.72% similarVerifiedPrompt injection attacks fundamentally exploit the inherent flaw of large models in failing to separate instructions from data, and are divided into two categories: direct injection and indirect injection.71% similarUnverifiedTools calling (Function Calling) allows a model to actively trigger predefined functions during a conversation and incorporate the results into its response, and was released by large model vendors earlier than MCP.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1030548API
curl https://kongchang.com/api/v1/knowledge/claims/1030548MCP
get_claim(id=1030548)