可证伪的安全宣称：把复核协议写进准入门槛

摘要

不可证伪的安全宣称在认识论上是空的。本文主张，进入高后果场景的系统，其安全宣称必须附带可证伪的复核协议——明确说明在什么条件下第三方可判定其为假——并把这种可证伪性本身写进准入门槛。我们给出与 ISA 形成证据相衔接的政策实现路径。

Abstract

An unfalsifiable safety claim is epistemically empty. We argue that systems entering high-consequence settings must attach a falsifiable verification protocol to their safety claims, and that this falsifiability itself be written into admission thresholds. We outline a policy implementation path connected to ISA formation evidence.

BibTeX

@techreport{trantorlabs2026falsifiable,
  title  = {Falsifiable Safety Claims},
  author = {TRANTOR LABS Research Group},
  year   = {2026},
  institution = {TRANTOR LABS}
}