Secure Formal Raising：用可执行属性保障 AI 开发过程安全

Secure Formal Raising: Protecting the Development of AI through Executable Properties

AI 导读

Secure Formal Raising（SFR）将 Formal Raising 扩展到 raising 环境的安全层面，用四值证据建模策略完整性、授权与机密数据流，保留未知和不一致评估。作者证明了六条安全不变量，并表明这些不变量迫使准入只接受已确认证据。一个纯 C 参考实现在有界穷举探索中满足这些不变量，同时暴露出工具调用中的 confused-deputy 缺口，可通过来源要求关闭。

正文 · 原文

View PDF HTML (experimental)

Abstract:Model raising revisits Turing's developmental view of machine intelligence. Formal Raising (FR) has recently been proposed to complement this view with explicit behavioral properties. This paper extends that proposal to the security of the raising environment. Secure Formal Raising (SFR) models policy integrity, authorization, and confidential data flow using four-valued evidence that preserves unknown and inconsistent assessments. We prove six security invariants and show that they force admission to accept only confirmed evidence. A plain-C reference implementation satisfies the invariants in a bounded exhaustive exploration, which also exposes a confused-deputy gap for tool calls that a provenance requirement closes.

Subjects:	Artificial Intelligence (cs.AI)
Cite as:	arXiv:2611.08042 [cs.AI]
	(or arXiv:2611.08042v1 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2611.08042 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Christina Plump [view email]
[v1] Thu, 5 Nov 2026 13:59:27 UTC (20 KB)

来源：arXiv cs.AI · arxiv.org