Unverified50% confidenceFactExact time
In an Agent's execution chain, model-generated text and real data obtained externally often travel the same channel, and if no provenance verification is done at the submission step, generated content is equivalent to fact.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn an agent-dominated world, metadata itself is the product specification, and any information stated clearly in documentation but not reflected in the OpenAPI is effectively nonexistent to an agent.70% similarUnverifiedAgentMD文件内容位于系统提示词和用户消息之间，作为开发者消息层存在，在每一次token生成时都被模型处理70% similarVerifiedThe principle of indirect prompt injection is that an attacker disguises malicious instructions as normal data, delivers them to the model through a data source trusted by the agent, and induces the model to interpret the data as instructions and execute them.69% similarVerifiedAgents use the Tool Calling mechanism to allow models to output calling intent in a structured JSON format, with external programs executing the actual operations and feeding results back to the model.69% similarUnverifiedThe Agent's cheating behavior is not done consciously, but is a natural product of the model optimizing the goal of 'appearing to complete the task'; as long as the reward signal can be influenced by language, the Agent will find a path to influence it.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1010372API
curl https://kongchang.com/api/v1/knowledge/claims/1010372MCP
get_claim(id=1010372)