reddit · SideProject · 10-08 18:50 61 分 ← 返回列表

I built an open-source tool that creates a "proof of custody" record for every AI answer, to help with EU AI Act compliance.

Hello guys! I'm a developer who got frustrated with the "black box" problem in AI, especially for high-stakes decisions like loans or insurance. The EU AI Act is coming (Dec 2027), and it will require companies to log who handled an AI's answer and how much to trust each step . I couldn't find a tool that did this in a tamper-evident way, so I built one. What it does: It tags an AI answer and grades each step in the chain (source, model, tool, agent, human). The weakest link se

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：AI 高风险决策缺乏可审计、防篡改的“保管链”记录，EU AI Act 合规要求企业证明谁处理了 AI 答案及每一步可信度。

目标用户：EU 受监管行业的合规官、DPO、AI 治理负责人，尤其是金融、保险、信贷等高风险 AI 场景。

证据：The EU AI Act is coming (Dec 2027), and it will require companies to log who handled an AI's answer and how much to trust each step. / Compliance officers and DPOs who need to hand an auditor a folder, not a messy log file.

变现：Open-core：自托管免费，托管 SaaS 按调用量/坐席收费；企业版提供 SSO/RBAC、审计报告模板、长期留存、SLA、合规集成与咨询。

MVP：2 周内做一个面向单一高风险场景的合规记录 MVP：SDK/API 接收 AI 答案元数据，生成哈希链与 Merkle 日志，签名收据，并一键导出审计用 PDF/JSON 包。

竞品：Credo AI、Holistic AI、IBM watsonx.governance、Fiddler AI、Arthur AI、WhyLabs

摘要：开发者发布 Apache-2.0 开源工具，为 AI 答案创建防篡改的“proof of custody”记录，按 source/model/tool/agent/human 分级并用 SHA-256、HMAC/Ed25519、Merkle log 密封。目标用户是 EU AI Act（2027 年 12 月）下的合规官和 DPO，需要向审计员提交文件夹而非杂乱日志。帖子没有收入或明确付费意愿证据，但监管截止日期带来潜在的企业托管与合规服务变现机会。