Unverified50% confidenceFactExact time
A well-aligned system with insufficient robustness may be bypassed its safety restrictions by carefully constructed prompts.
1
Sources
50%
Confidence
Long-term
Relevance
10/29/2026
First Seen
Sources
Related Entities
Related Claims
Unverified"No-crash property" is a safety property that only constrains that "bad things do not happen" and does not require that "good things must happen"; by contrast, functional correctness requires complete input-output specifications, and its proof difficulty is an order of magnitude higher.78% similarUnverified纵深防御原则要求系统安全性不应依赖于用户的完美操作，每一层工具都应假设上一层可能已被突破并提供独立的安全校验78% similarUnverifiedSetting the security boundary outside the prompt is not a distrust of model capabilities, but a respect for the basic laws of systems engineering.77% similarVerifiedPrompt-level guardrails (constraining agent behavior solely through system prompts) are inherently fragile, and a more reliable direction is to enforce controls at the tool-call level.75% similarUnverifiedThe reliable defense against prompt injection is architectural least-privilege isolation: physically separating the steps that process untrusted content from the steps that have tool permissions, and adding a final human confirmation.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1112714API
curl https://kongchang.com/api/v1/knowledge/claims/1112714MCP
get_claim(id=1112714)