Unverified50% confidenceFactExact time
Secret trigger words are essentially a variant of backdoor attacks, in which hidden associations are implanted in the model during training or deployment, and the model's behavior is hijacked when the input contains a specific pattern.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBackdoor 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.79% similarVerified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% similarUnverifiedIndirect prompt injection is harder to defend against than direct injection because the attack payload is hidden within external content the model is supposed to trust, exploiting the fundamental characteristic of LLMs that mixes instructions with data.73% 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% 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% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1079582API
curl https://kongchang.com/api/v1/knowledge/claims/1079582MCP
get_claim(id=1079582)