English(EN) TPM-Bound AI Identity In most AI deployments the identity of the acting agent is a portable secret, and anything portable can be stolen or cloned. We bind ident

Mickai提出主权AI，采用可验证的、离线优先架构

作者 PulseAugur 编辑部 · [10 个来源] · 2026-07-07 03:07

Mickai正在提出一个“主权AI”框架，强调可验证的操作和数据所有权，摆脱对云的依赖。该方法包括使用签名记录、使用SHA-3-512构建的防篡改审计链，以及ML-DSA-65等后量子密码学进行签名。该架构旨在实现离线优先运行，通过受信任平台模块（TPM）将AI身份绑定到硬件，以确保可防御性和可审计性，即使面对未来的量子威胁。 AI

影响该框架可能通过优先考虑可验证的操作和数据主权来影响未来的AI发展，可能带来更具可审计性和更安全的AI系统。

排序理由该集群包含来自Mastodon上同一作者的多个博客文章，讨论了AI架构和治理的概念框架，而不是产品发布或研究论文。

在 Mastodon — mastodon.social 阅读 →

AI 生成摘要 · Google Gemini · 来自 10 个来源。我们如何撰写摘要 →

报道来源 [10]

Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:12

董事会与人工智能监督职责董事会只能监督他们能验证的内容，而生产中的大多数人工智能都无法进行检查。签名，哈希链接行

The Board and the AI Oversight Duty Directors can only oversee what they can verify, and most AI in production offers nothing to check. Signed, hash-linked lineage on hardware the board controls is what turns a governance slide into a governance fact. https:// mickai.co.uk/articl…

链接 mickai.co.uk/…/the-board-ai-oversight-duty
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:11

CIO主权人工智能路线图：构建主权智能资产并非一次性采购，而是分阶段的旅程，从依赖云到拥有自己的智能

The CIO Sovereign AI Roadmap Reaching a sovereign intelligence estate is not a single purchase but a staged voyage, from cloud dependency to intelligence you own, govern and can defend under any audit. We walk the CIO through each stage, using the journey of Odysseus as our map. …

链接 mickai.co.uk/…/the-cio-sovereign-ai-roadm…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:11

主权式AI的首席法务官案例：披露和可辩护性是首席法务官向企业AI提出的三个问题。主权式

The General Counsel Case for Sovereign AI Privilege, disclosure and defensibility are the three questions every general counsel asks of enterprise AI. Sovereign, signed and owned records answer all three. https:// mickai.co.uk/articles/the-gene ral-counsel-case-for-sovereign-ai #…

链接 mickai.co.uk/…/the-general-counsel-case-f…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:10

AI模型的数字水印和溯源溯源不是模型上的贴纸，而是贯穿整个布料的线。Mickai 如何绑定可验证的...

Watermarking and Provenance for AI Models Provenance is not a sticker on the model, it is thread woven through the whole cloth. How Mickai binds verifiable, attributable origin into every brain and every artefact, signed before it acts and verifiable offline. https:// mickai.co.u…

链接 mickai.co.uk/…/ai-model-watermarking-and-…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:09

零信任AI架构零信任抛弃边界，要求每一步都提供证明。以下是签名身份、每次操作的证明以及防篡改的...

Zero-Trust AI Architecture Zero trust discards the perimeter and demands proof at every step. Here is how signed identity, per-action attestation and a tamper-evident ledger turn trust nothing, verify everything, into working AI architecture. https:// mickai.co.uk/articles/zero-t…

链接 mickai.co.uk/…/zero-trust-ai-architecture
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:09

离线优先AI架构Antaeus 触地则无敌，离地则脆弱。这就是离线优先艺术的全部论点

Offline-First AI Architecture Antaeus was invincible while he touched the earth and weak the moment he left it. That is the whole argument for offline-first artificial intelligence: full-power sovereign intelligence that draws its strength from the operator's own ground, feeds on…

链接 mickai.co.uk/…/offline-first-ai-architect…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:08

签名记录：Mickai 内部的 CBOR 日志格式，没有任何操作是可信的，因为系统声称它发生了。它是可信的，因为紧凑、签名的 CBOR 记录

Signed Records: The CBOR Ledger Format Inside Mickai, no action is trusted because a system claims it happened. It is trusted because a compact, signed CBOR record proves it happened, in this order, with this outcome, and can be replayed and verified offline years later. https://…

链接 mickai.co.uk/…/signed-records-the-cbor-le…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:08

SHA-3-512 和防篡改审计链基于SHA-3-512的哈希链接审计账本使篡改过去在数学上不可能：每个条目都带有i

SHA-3-512 and Tamper-Evident Audit Chains A hash-linked audit ledger built on SHA-3-512 makes altering the past mathematically impossible: every entry carries its whole ancestry, so the smallest edit shatters the chain and names the record where it broke. https:// mickai.co.uk/ar…

链接 mickai.co.uk/…/sha3-512-tamper-evident-au…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:08

ML-DSA-65：每个动作之下的签名每个动作之下，一个Mickai大脑运行着一个简单而安静的机制：一个后量子FIPS 204 ML-DSA-65签名

ML-DSA-65: The Signature Under Every Action Beneath every action a Mickai brain takes sits a single quiet mechanism: a post-quantum FIPS 204 ML-DSA-65 signature that turns an operation into evidence built to hold for decades, even against the quantum machines of a future we canno…

链接 mickai.co.uk/…/ml-dsa-65-the-signature-un…
Mastodon — mastodon.social TIER_1 English(EN) · Mickai · 2026-07-07 03:07

TPM 绑定的 AI 身份在大多数 AI 部署中，运行代理的身份是一个可移植的秘密，任何可移植的东西都可能被盗窃或克隆。我们将身份绑定

TPM-Bound AI Identity In most AI deployments the identity of the acting agent is a portable secret, and anything portable can be stolen or cloned. We bind identity and keys to the operator's own TPM, so only the right party, on the right machine, may cross. https:// mickai.co.uk/…

链接 mickai.co.uk/…/tpm-bound-ai-identity

报道来源 [10]

相关实体

相关话题