Unverified60% confidenceFactExact time
Backdoor attacks plant triggers in training data, causing the model to behave normally on regular inputs but produce preset malicious outputs when specific trigger words are encountered.
2
Sources
60%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
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% similarUnverifiedSensitive information from training data can be extracted by reverse-engineering model weights, which is known as a model inversion attack.75% similarUnverifiedModel behavior is shaped jointly by training data, prompts, context, and external tool calls, and contamination or misuse of any one link can lead to harmful output, leakage of sensitive information, or unauthorized operations.74% similarUnverifiedAfter alignment training eliminates a certain type of harmful output, the model may learn more subtle ways of expression.74% 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.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/501426API
curl https://kongchang.com/api/v1/knowledge/claims/501426MCP
get_claim(id=501426)