Unverified50% confidenceFactExact time
Model poisoning can be classified by attack stage into training-time poisoning and deployment-time poisoning; what NemoClaw demonstrates is deployment-time poisoning—directly tampering with the weight parameters or configuration of an already-deployed model.
1
Sources
50%
Confidence
Long-term
Relevance
9/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedModel poisoning involves implanting backdoors or biases during the training phase of an AI model, causing the model to produce erroneous judgments under specific conditions.74% similarUnverifiedBackdoor 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.66% 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.66% similarUnverifiedAfter alignment training eliminates a certain type of harmful output, the model may learn more subtle ways of expression.63% similarUnverified基于LLM的系统天然面临提示词注入（Prompt Injection）攻击62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/868422API
curl https://kongchang.com/api/v1/knowledge/claims/868422MCP
get_claim(id=868422)