Tag
提出 Structured Authority Injection (SAI) 协议,通过带类型的 MCP 覆盖调用和权威上下文注入,使人类操作员能对 LLM 智能体实施跨规划周期的持久控制,无需微调模型。
QuixiAI 开源了 Sovereign Protocol Language (SPL),一个基于 Lean 4 的形式化库,为"主权代理协议"(Protocol of Sovereign Agency) 提供规范性语义,核心包括 Duty、Power、Immunity 等原语以及 Law 3 非干扰实验,全部定理经内核验证、无公理依赖。
作者批评 ML-DSA 和 Ed25519 在签名时先对公钥哈希的做法,认为这破坏了验证端先哈希消息再逐个试验签名密钥的高效流程,并建议把公钥哈希拼接在消息末尾以保留流式哈希的优势。
Agent Network Protocol (ANP) is an open-source protocol suite positioning itself as the HTTP of the agentic web, covering agent identity, naming, discovery, negotiation, and secure messaging; the post highlights a design distinguishing approved service identity from explicit, auditable delegation with revocation checks.
The article advocates for mesh networks to use message signing instead of encryption in amateur radio, highlighting benefits like authenticity and easier legal operation, while noting drawbacks such as reduced privacy.
AIREP is a protocol for recording per-decision governance evidence in AI runtimes, providing transparency and accountability by creating signed, tamper-evident records of decisions under stated policies.
The author redesigned the IONS protocol from brute-force graph traversal to a Cognitive Attention Architecture that progressively routes queries through relevant slices of the network, separating path confidence, relevance, and utility.
This paper introduces Preregistered Belief Revision Contracts (PBRC), a protocol-level mechanism for multi-agent systems (including LLM-based agents) that separates open communication from admissible belief changes by publicly fixing evidence triggers and revision operators. The work addresses dangerous conformity effects in agent deliberation and provides formal guarantees that social-only pressure cannot drive false consensus.