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

博客园 - 三生石上(FineUI控件)
S
SegmentFault 最新的问题
Jina AI
Jina AI
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
IT之家
IT之家
美团技术团队
V
Visual Studio Blog
人人都是产品经理
人人都是产品经理
博客园 - 叶小钗
有赞技术团队
有赞技术团队
GbyAI
GbyAI
宝玉的分享
宝玉的分享
腾讯CDC
M
MIT News - Artificial intelligence
博客园 - 【当耐特】
Google DeepMind News
Google DeepMind News
月光博客
月光博客
MyScale Blog
MyScale Blog
Last Week in AI
Last Week in AI
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
博客园 - 司徒正美
Recent Announcements
Recent Announcements
MongoDB | Blog
MongoDB | Blog

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 - radotsvetkov/akmon: Tamper-evident evidence and ...
radotsvetkov · 2026-05-10 · via Hacker News - Newest: "AI"
            ✦        ✦        ✦

           ▓▓▓
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
         ▒▒▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▒▒
         ▒▒▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▒▒
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
             ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
               ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
                   ▓▓▓▓▓▓▓▓▓▓▓▓
                    ▓▓      ▓▓
                    ▓▓      ▓▓
                 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
               ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
             ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓

Akmon is a tamper-evident evidence and verification layer for AI agents. It sits on top of whatever agent you already run, whether that is your own or any tool that emits OpenTelemetry. Each session becomes a portable, content-addressed, cryptographically signed record that someone else can verify for themselves. The part that matters most is this: a third party can check a signature offline using nothing but openssl, with no Akmon install, no cloud service, and no need to trust whoever produced the record.

Website: radotsvetkov.github.io/akmon. Documentation: radotsvetkov.github.io/akmon/docs.

CI Passed tests License Rust 1.88+

Why this exists

When an AI agent changes something, you may have to prove later what it actually did. The person asking could be a regulator, a security reviewer, or an incident team, and it might be years after the fact. They may not trust you, and they may not run your tools. Most agent telemetry cannot stand up to that. It lives in process memory, or in mutable, unsigned spans, and "the AI did it" is not an answer anyone will accept. Under the EU AI Act, the high-risk logging obligations in Article 12 and Annex IV start to apply on 2 August 2026.

Akmon treats the evidence itself as the thing you ship. It takes a session, either your own or one from any OpenTelemetry-instrumented agent, and turns it into a sealed record. Someone else can then check that record's integrity and authorship independently, with standard tools, even on a machine that has never heard of Akmon.

The trust toolchain

Every command here ships today and is covered by tests. The focus is the verification chain, not the agent.

Step Command What it does
Import any agent akmon otel import <trace.json> Turns an OpenTelemetry GenAI trace into an AGEF session. It reads the current v1.37 structured form and the older v1.36 message-event form that most agents still emit. The capture level (full or structural) is recorded honestly.
Generate a key akmon bundle keygen --out k.pk8 --public-out k.pub Creates an Ed25519 signing key (PKCS#8 v2).
Sign akmon bundle sign <bundle> --key k.pk8 Adds an offline Ed25519 signature over the session head.
Attest an operator akmon bundle attest <bundle> --key op.pk8 --operator-id you@org --role approver Records the accountable person behind a session, signed separately from the head signature.
Verify akmon bundle verify <bundle> --verify-key k.pub --require-signature Checks integrity, the signature, any operator attestation, and the capture level.
Verify on its own agef-verify <bundle> --verify-key k.pub A small, separate binary for auditors. It does not need the full Akmon install.
Prove with openssl akmon bundle prove-openssl <bundle> --verify-key k.pub --out-dir proof Writes out the statement, signature, and public key so anyone can check the signature with plain openssl.

Other commands round out the lifecycle: bundle export and bundle import, inspect, diff, replay (deterministic playback of sessions Akmon produced itself), redact, audit, evidence, verify, policy, and doctor.

Quickstart

You can install Akmon a few ways.

On macOS or Linux with Homebrew:

brew tap radotsvetkov/akmon
brew install akmon         # the CLI
brew install agef-verify   # the standalone verifier for auditors

Or grab the prebuilt binaries directly. Each GitHub release attaches prebuilt akmon and agef-verify binaries for Linux and macOS, plus a SHA256SUMS file so you can check what you downloaded. For Linux on x86_64:

base=https://github.com/radotsvetkov/akmon/releases/latest/download
curl -LO $base/akmon-linux-x86_64
curl -LO $base/agef-verify-linux-x86_64
curl -LO $base/SHA256SUMS

# Check the downloads against the published checksums before installing.
sha256sum --check --ignore-missing SHA256SUMS

chmod +x akmon-linux-x86_64 agef-verify-linux-x86_64
mv akmon-linux-x86_64 ~/bin/akmon
mv agef-verify-linux-x86_64 ~/bin/agef-verify

On macOS the file names are akmon-darwin-arm64 and agef-verify-darwin-arm64 for Apple silicon, or the -x86_64 variants for Intel, and you check them with shasum -a 256 --check --ignore-missing SHA256SUMS.

Or build from source on any platform:

cargo install --git https://github.com/radotsvetkov/akmon akmon
cargo install --git https://github.com/radotsvetkov/akmon agef-verify

Here is a full run, from any agent's OpenTelemetry trace to a proof a stranger can check.

# 1. Make a signing key. Keep the private half secret and hand out only the .pub.
akmon bundle keygen --out signer.pk8 --public-out signer.pub

# 2. Turn a real trace into a session, then a portable bundle.
akmon otel import trace.json --journal ./journal
akmon bundle export <session-id> --journal ./journal --output audit.akmon

# 3. Sign it.
akmon bundle sign audit.akmon --key signer.pk8

# 4. Verify integrity, the signature, and the capture level.
akmon bundle verify audit.akmon --verify-key signer.pub --require-signature

# 5. Prove it with openssl alone, no Akmon involved.
akmon bundle prove-openssl audit.akmon --verify-key signer.pub --out-dir proof
openssl pkeyutl -verify -pubin -inkey proof/pubkey.pem -rawin -in proof/statement.bin -sigfile proof/signature.bin

Use OpenSSL 3.x for the verify step. The openssl that ships with macOS is LibreSSL, which cannot verify Ed25519. Install OpenSSL 3.x first, for example with brew install openssl@3.

There is a full walkthrough in the docs: from an OTEL trace to an offline openssl proof.

How much a session captures

Akmon is explicit about how complete each record is, and it signs that level into the record so nobody can quietly overstate it.

Full capture (capture_level = full) comes from Akmon's own agent. It keeps the prompts, responses, and tool calls, so the session replays deterministically with akmon replay.

Structural capture (capture_level = structural) comes from importing another agent's OpenTelemetry trace. You get the shape of the session, its provider calls, tool calls, metadata, and whatever content the trace happened to include, but not a byte-for-byte replay. akmon bundle verify --require-capture full fails on these on purpose, and akmon replay refuses them.

An imported trace is never dressed up as a full recording. Keeping that line clear is the whole reason a trust layer is worth having.

How it compares to Microsoft

Microsoft ships a solid governance runtime, the open-source Agent Governance Toolkit, which has been generally available since April 2026, and a strong tamper-evident cloud ledger in Azure Confidential Ledger. As of June 2026, though, no single Microsoft product gives you one portable, self-contained evidence artifact that is signed with an asymmetric key (Ed25519), checkable by a stranger who has nothing from Microsoft installed, deterministically replayable, and able to sit on top of any agent. The Toolkit's tamper-evidence is a hash chain plus HMAC, with no asymmetric signature and no standalone verifier. Confidential Ledger's signed Merkle receipts are genuinely good, but they are tied to Azure and are not aware of agents. Microsoft's own Foundry documentation says its traces cannot support full replay.

That gap is where Akmon fits. It is portable, signed, cloud-independent, agent-aware, and replayable.

Akmon is not trying to replace any of that, and there are places where Microsoft is clearly ahead. Its distribution is one: Purview and the Copilot Control System are already in every Microsoft 365 tenant. Confidential Ledger's offline-verifiable receipts are another, and so is Microsoft's weight in the standards bodies. A single tool will not match those. Akmon is meant to complement them: seal what Purview captures, and export and verify what Foundry traces.

Compliance

Akmon is built to help you produce evidence for frameworks like the EU AI Act (Article 12 and Annex IV logging, with high-risk obligations from 2 August 2026), the NIST AI Risk Management Framework (MEASURE 2.8), and SOC 2 (CC7.x and CC8.1). It is not a compliance certification, and it does not guarantee compliance. Treat any mapping from AGEF evidence to specific controls as something to validate with your own compliance and legal teams.

The bundled agent

Akmon ships with its own agent (akmon, akmon chat, and akmon --task ...). It has typed permission checks for file writes, shell, and network, runs local or hosted models, supports policy profiles, and governs MCP servers. It exists mainly as the reference producer, the way to get full-capture sessions that replay deterministically. It is not trying to out-code Cursor or Claude Code. The value Akmon offers is the evidence layer, and that works no matter which agent you prefer.

# A session from Akmon's own agent, with its verification pipeline.
akmon --yes --output json --task "apply rustfmt fixes only" | tee run.json
akmon audit verify .akmon/audit/<session-id>.jsonl
akmon evidence verify .akmon/evidence/<session-id>.json
akmon slo verify .akmon/evidence/<session-id>.json --strict

The evidence format (AGEF)

Akmon's records follow the AGEF specification, a portable, content-addressed, tamper-evident format for AI-agent session evidence. Akmon implements AGEF v0.1.3. That version adds two optional pieces on top of the hash chain: offline Ed25519 signatures in manifest.signatures[], which make a record attributable to a key, and operator attestations in manifest.operator_attestations[], which record the accountable person behind a session. Both are optional, so a plain bundle stays small and older readers keep working.

Documentation

Contributing

License

Apache-2.0 only. See LICENSE.

Where the name comes from

Akmon is an old Greek word for anvil. The idea is to shape difficult work with pressure and precision while keeping control over every strike: the permissions, the tamper-evident audit trail, and the evidence anyone can verify.