惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

量子位
博客园_首页
罗磊的独立博客
云风的 BLOG
云风的 BLOG
J
Java Code Geeks
Last Week in AI
Last Week in AI
D
DataBreaches.Net
Jina AI
Jina AI
博客园 - Franky
大猫的无限游戏
大猫的无限游戏
Apple Machine Learning Research
Apple Machine Learning Research
V
V2EX
D
Docker
MongoDB | Blog
MongoDB | Blog
B
Blog RSS Feed
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
宝玉的分享
宝玉的分享
Engineering at Meta
Engineering at Meta
The Cloudflare Blog
博客园 - 三生石上(FineUI控件)
有赞技术团队
有赞技术团队
人人都是产品经理
人人都是产品经理
H
Help Net Security
T
The Blog of Author Tim Ferriss

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - aerf-spec/aerf: Agent Evidence Receipt Format (A...
keertahacker · 2026-05-07 · via Hacker News - Newest: "AI"

v0.1.0-draft.1 — Public Review Draft, May 2026. Not yet stable.

For reviewers:

Goal Link
All framework coverage at a glance docs/COMPLIANCE.md
AIUC-1 mapping (primary target) docs/frameworks/AIUC-1.md
Run the verifier § Try it
Receipt example § Sample receipt
Specification SPEC.md

AERF is an open wire format for cryptographic receipts of AI-agent actions. Each receipt is an Ed25519-signed JSON document that records what an agent did, what policy permitted it, when it happened, and the full evidence of the action. Receipts are independently verifiable — no AERF software, account, or service is needed to check one. The reference verifier in this repo is a single Go file using only the standard library.

AERF addresses the evidence and logging layer of AI governance frameworks. It supplies tamper-evidence and independent-verifiability evidence for controls in AIUC-1, HIPAA, SOC 2, ISO/IEC 42001, the EU AI Act, NIST AI RMF, SR 11-7, and SOX 404. AERF does not replace full compliance programs; see docs/COMPLIANCE.md for the per-control mapping and explicit gaps.

This repository is the home of the specification and a reference verifier. The reference producer lives in agentmint-python (pip install agentmint).

AERF is to agentic AI evidence what cosign is to container images and slsa-verifier is to build provenance: a small, boring, auditable file format with a small, boring, auditable verifier.

Sample receipt

{
  "id": "7473e179-001c-4d3b-94bc-d0f53dd6eec6",
  "type": "notarised_evidence",
  "plan_id": "bc023208-ea24-410a-a280-ff36820e18a6",
  "agent": "claims-agent",
  "action": "submit:claim:CLM-9920",
  "in_policy": true,
  "policy_reason": "matched scope submit:claim:*",
  "evidence_hash_sha512": "b35d8ba27ad113c4...bb39e30c",
  "evidence": { "...": "..." },
  "observed_at": "2026-05-06T16:22:33.490443+00:00",
  "policy_hash": "260eca8ac43ae65e...985d6bf1",
  "key_id": "c348d3c785c92249",
  "plan_signature": "3e5b83e83b77bfa2...dea8ee01",
  "signature": "8bd989a95ab60863...04e97208"
}

evidence field abbreviated for display; see verifiers/go/example/receipt.json for the full file. evidence_hash_sha512, policy_hash, plan_signature, and signature truncated.

The signature is an Ed25519 signature over the canonical JSON encoding of the receipt with signature and timestamp removed. A receipt by itself plus an issuer's public key is a complete audit artifact.

Try it

git clone https://github.com/aerf-spec/aerf.git
cd aerf/verifiers/go
go run verify.go example/receipt.json example/public_key.pem
# → "OK  receipt 7473e179..." and exit code 0

# Confirm tamper detection
go run verify.go example/receipt-tampered.json example/public_key.pem
# → "FAIL signature verification FAILED ..." and exit code 1

The tampered file differs from the original by a single field (CLM-9920CLM-9921). The signature does not.

What's in this repo

.
├── README.md                          You are here.
├── SPEC.md                            The AERF-EVIDENCE specification (draft).
├── DECISIONS.md                       Locked + held design decisions C-1..C-20.
├── CHANGELOG.md
├── LICENSE                            Apache 2.0 — code, schemas, examples.
├── LICENSE-spec                       CC BY 4.0 — prose / specification text.
├── docs/
│   ├── COMPLIANCE.md                  Compliance navigation hub.
│   └── frameworks/                    Per-framework AERF mapping pages.
├── schemas/
│   └── aerf-v0.1.json                 JSON Schema (Draft 2020-12) for the
│                                      EVIDENCE receipt shape.
└── verifiers/
    └── go/
        ├── verify.go                  ~200 lines, stdlib only.
        ├── go.mod
        ├── README.md
        └── example/
            ├── receipt.json
            ├── receipt-tampered.json
            ├── public_key.pem
            └── evidence-package.zip   Full agentmint evidence ZIP.

What's NOT in v0.1.0-draft.1

These are intentionally deferred. They will land in subsequent drafts:

  • Test vectors — a directory of ~8 conformance vectors (genesis, chain, tamper, replay, malformed, etc.).
  • Python and TypeScript reference verifiers. Today, Python is the reference producer; only the Go verifier ships in this repo.
  • compliance/ as a normative directory — superseded by the non-normative docs/ framework mapping pages (AIUC-1, HIPAA, SOC 2, ISO/IEC 42001, EU AI Act, NIST AI RMF, SR 11-7, SOX 404). A normative compliance/ directory under spec governance (locked decision C-20) remains deferred.
  • Governance, contributing, and security policy documents.
  • CI workflows and pre-built release binaries of the verifier.
  • AERF-AUTHZ profile — the spec acknowledges it as a future profile (held decision C-17) but does not specify it. v0.1 ships the EVIDENCE profile only.
  • Reference verifier hash-chain and RFC 3161 timestamp checks. Both are described normatively in the spec; the Go reference verifier in this draft enforces signatures only. See verifiers/go/README.md.

Status and stability

This is a public review draft. The wire format may change before v0.1.0 stable. Locked decisions (see DECISIONS.md) are binding for v0.1.0; held decisions remain open until v0.1.0 stable. We tag every page header and the spec title page accordingly so nothing here gets cited as final.

Reference implementation

The reference producer at the time of this draft (agentmint 0.1.x) diverges from the spec on two locked decisions (genesis sentinel C-6 and chain hash input C-7). The v0.1.0-draft.1 example is a single genesis receipt to sidestep the gap; library fixes are tracked in issue #2 for v0.1.0-draft.2.

Insurance carriers can use Plan receipts as machine-readable underwriting intake and Evidence receipts as tamper-evident claims evidence. See Appendix B.

License

Dual-licensed:

  • Prose (SPEC, README, DECISIONS, CHANGELOG) — Creative Commons Attribution 4.0 International (LICENSE-spec).
  • Code, schemas, example artifacts (verifiers, schemas, receipt.json, etc.) — Apache License 2.0 (LICENSE).

If a file's license is unclear, the per-directory README.md or the file's own header governs.

Editor

Aniketh Maddipati (agentmint.run). Issues and PRs welcome.