Unverified60% confidenceTradeoffExact time
Vague tool authorization prompt descriptions can easily trigger completely different execution strategies in different models.
2
Sources
60%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAgent behavior is highly non-deterministic, and the same input may produce completely different tool call chains across different runs.77% similarUnverifiedAbliteration is implemented by directly modifying model weights, which differs from bypassing safety restrictions at inference time via system prompts.76% similarUnverifiedMost bizarre agent behaviors are rooted in poor tool descriptions, because the model guesses meaning from descriptions rather than reading code logic.74% similarUnverifiedThe least-privilege agent design pattern decouples the model's reasoning capabilities from its tool execution capabilities and introduces an independent policy enforcement layer between the two.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/932490API
curl https://kongchang.com/api/v1/knowledge/claims/932490MCP
get_claim(id=932490)