Unverified50% confidenceFactExact time
Models are trained to do their best to satisfy user requests, while enterprise policies often require refusing requests under specific conditions, and when boundaries are not explicitly encoded into the system the model tends to choose to help.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDo not rely on a small decision model as the sole authority for 'whether the user requested this call'; small models will say yes to calls that should be rejected, and what actually works is provenance and hard policies in code.73% similarUnverifiedEngineering practices based on the assumption that a model has been compromised include: independently validating model outputs, applying the principle of least privilege to restrict the data and tools accessible to the model, continuously monitoring and auditing inputs and outputs, and isolating scenarios that involve ingesting external content.72% similarUnverifiedThe model routing system dynamically selects the most suitable underlying model based on task type, complexity, cost or latency requirements, which can avoid single-model-vendor lock-in but brings management issues such as consistency of model behavior, differences in output formats, and cross-model security and compliance.70% similarUnverifiedSystem prompt adherence is not a hard constraint but a probabilistic tendency; models may ignore system instructions when user input creates a semantic conflict with the system prompt.70% similarUnverifiedThe model never directly touches databases, APIs, or files; it issues structured tool requests, which the orchestrator validates, executes, and returns results for, and the model can only reach the outside world through the capabilities the application chooses to expose.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1030481API
curl https://kongchang.com/api/v1/knowledge/claims/1030481MCP
get_claim(id=1030481)