Unverified50% confidenceFactExact time
In LLM agent scenarios, the model treats tool description fields as part of its own configuration to follow and treats result fields as external data to process, and this difference in trust level causes the same text to carry different injection risk in different fields.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIndirect 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.75% similarUnverifiedThe LLM judge model inherits the biases in its own training data and may systematically prefer a certain tone, length, or style of expression, so the audit framework records the judge model name as a separate field to achieve transparency.75% similarUnverifiedThe paper has LLMs interpret the textual narrative of each variant in the context of the goal model and assign it to the most appropriate category, using the LLM's semantic reasoning capability to connect low-level process behavior with high-level business objectives.73% similarUnverifiedThe LLM agent's evidence loop depends on two capabilities: tool orchestration (knowing when to call which reverse-engineering tool and what parameters to pass) and semantic understanding (reading the design intent behind the pseudocode).71% similarUnverifiedLLM在表格场景中更适合承担辅助角色，如特征工程、解释预测结果、处理文本字段、数据清洗，最终预测建模仍由树模型承担71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1057945API
curl https://kongchang.com/api/v1/knowledge/claims/1057945MCP
get_claim(id=1057945)