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

aimingoo的专栏
aimingoo的专栏
Google DeepMind News
Google DeepMind News
S
SegmentFault 最新的问题
Project Zero
Project Zero
D
DataBreaches.Net
I
InfoQ
L
Lohrmann on Cybersecurity
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
酷 壳 – CoolShell
酷 壳 – CoolShell
Stack Overflow Blog
Stack Overflow Blog
The Register - Security
The Register - Security
Recorded Future
Recorded Future
Vercel News
Vercel News
博客园 - 司徒正美
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
I
Intezer
The Hacker News
The Hacker News
F
Fortinet All Blogs
Microsoft Azure Blog
Microsoft Azure Blog
P
Proofpoint News Feed
Help Net Security
Help Net Security
cs.CV updates on arXiv.org
cs.CV updates on arXiv.org
Application and Cybersecurity Blog
Application and Cybersecurity Blog
Scott Helme
Scott Helme
T
Threatpost
爱范儿
爱范儿
N
Netflix TechBlog - Medium
D
Docker
云风的 BLOG
云风的 BLOG
C
Cisco Blogs
K
Kaspersky official blog
H
Help Net Security
S
Secure Thoughts
T
Threat Research - Cisco Blogs
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
S
Security @ Cisco Blogs
Cyberwarzone
Cyberwarzone
N
News and Events Feed by Topic
G
Google Developers Blog
Forbes - Security
Forbes - Security
博客园 - 三生石上(FineUI控件)
博客园 - 叶小钗
B
Blog
Google DeepMind News
Google DeepMind News
Recent Announcements
Recent Announcements
Simon Willison's Weblog
Simon Willison's Weblog
S
Securelist
P
Privacy International News Feed
Spread Privacy
Spread Privacy
The Last Watchdog
The Last Watchdog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant Common SOC 2 Failures (Real World) Stop Vibe-Checking Your AI App: A Practical Guide to Evals How to Use SonarQube and SonarScanner Locally to Level Up Your Code Quality Your Next To-Do App Is Dead — I Replaced Mine with an OpenClaw AI Sign a Nostr event in 60 lines of Python using coincurve — no nostr-sdk, no nbxplorer, no rust toolchain ITGC Audit Explained Like You’re in Big 4 Patch Tuesday abril 2026: Microsoft parcha 163 vulnerabilidades y un zero-day en SharePoint Stop scraping everything: a better way to track competitor price changes Listing on MCPize + the Official MCP Registry while routing payments OUTSIDE the marketplace — how I kept 100% of my x402 revenue Building an AI-Powered Risk Intelligence System Using Serverless Architecture Why We Ripped Function Overloading Out of Our AI Toolchain Testing AI-Generated Code: How to Actually Know If It Works SaaS Churn Is Killing Your Business. Here Is What to Do About It (Without a Support Team) The Speed of AI Is No Longer Linear - And Self-Improving Models Are Why How to Implement RBAC for MCP Tools: A Practical Guide for Engineering Teams From Standard Quote to Persuasive Proposal: AI Automation for Arborists I built a CLI that scaffolds complete multi-tenant SaaS apps Axios CVE-2025–62718: The Silent SSRF Bug That Could Be Hiding in Your Node.js App Right Now The dashboard that ended our friendship Data Pipelines Explained Simply (and How to Build Them with Python) The Hidden Cost of AI Systems Nobody Talks About. undefined vs undeclared, and how typeof behaves Switching from file-based jobs to NATS/Kafka in Rust without changing code io_uring Adventures: Rust Servers That Love Syscalls Why Agentic AI is Killing the Traditional Database The POUR principles of web accessibility for developers and designers Quantum Neural Network 3D — A Deep Dive into Interactive WebGL Visualization How To Install Caveman In Codex On macOS And Windows Automation Pipeline Reliability: Why Your Workflow Breaks When Nobody Is Watching I Built an 'Open World' AI Coding Agent — It Works From ANY Folder From Freelancing to Product: A Tech Service Company's SaaS Transformation China's AI Giants: Adding Tencent Hunyuan & ByteDance Doubao to AI University (74 Providers) On the Vibe Coders and Their Lies clerk: Auto-Summarize Your Claude Code Sessions AI Weekly — 2026/04/10–04/17 | The Model Lockdown Is Here, but the Toolchain Is the Real Battleground AI 週報 — 2026/04/10–2026/04/17 模型封鎖潮來了,但工具鏈才是真戰場 Maybe this is how Open-Source apps are born... 🚀 Fine-Tune LLMs with LoRA and QLoRA: 2026 Guide tRPC v11 + Next.js App Router: End-to-End Type Safety Without the Boilerplate ShadCN UI in 2026: Why I Stopped Installing Component Libraries and Started Owning My Components SaaS Billing in React Server Components: Stripe + Supabase Without a Single `useEffect` Join our DEV Weekend Challenge — $1,000 in Prizes Across TEN winners! Submissions Due April 20 at 6:59 AM UTC. Implementing FSRS Spaced Repetition in Flutter + Supabase — Adding Memory Science to an AI Learning App "I Texted My Localhost From the Train — Claude Code Fixed the Bug Before I Got Home" I Built a Sales Prep AI and It Went Deeper Than Expected Design to Code #2: One JSON, Eleven Outputs Solving the 100M-Row Problem: A Summary Table Pattern for High-Volume Push Notification Logs Flutter Web With Wasm: What Actually Changes For Developers I Built 50 Royalty-Free Soundtracks for My Side Project in a Weekend Using AI Music Generation The Vibe Coding Security Checklist: 7 Things to Check Before You Ship Stop Letting Googlebot Guess Fix Your React App's SEO Right Desconstruindo o Streaming do LinkedIn: Como Criar um Engine de Extração de Vídeo de Alta Performance com HLS e FFmpeg (EDA Part-1) EDA (Exploratory Data Analysis) Explained With Real Life — Why Looking at Your Data Is the Most Important Step in Machine Learning Brand Relationship Management at Scale: Our 4-Touch Outreach System for 200+ Brands Why String.fromEnvironment() Might Return an Empty String in Dart JGuardrails 1.0.0 — Hardening Java LLM Apps Against Jailbreaks, Toxicity, and Prompt Injection Plan and Schedule a Full Week of Threads Content From One Claude Conversation Coding Cat Oran Ep3, Five Tables Changed Everything Updated: BFF Pattern I'm done watching freelancers get buried by 200 proposals. So I'm building the alternative. This is my first post BFS Algorithm in Java Step by Step Tutorial with Examples Tracking LLM Pricing Monthly: An Open Dataset for 22 AI Models How We Measure Content ROI on a Comparison Site: Revenue Attribution Without Perfect Data Introducing Nova AI Ops: The AI-Native Operating System for SRE Teams I built a free desktop video downloader for Windows — Grabbit How Talkie OCR Helps Vision-Impaired & Dyslexic Users Read the World Around Them VRCFaceTracking安装和iPhone面捕配置教程,有bug Even CrowdStrike Can't See Your Agents The Automation Gold Rush: What n8n Workflows and Claude Are Opening Up for Developers Right Now
Why Did Docker Abandon TUF?: A Turbulent History of Container Signing
kt · 2026-04-28 · via DEV Community

Introduction

While doing a deep dive on Sigstore and TUF, a question hit me out of nowhere.

"OK, but how exactly are container images protected from tampering?"

If you understand TUF, you'd guess: "You write the container image hash into targets.json, sign it with an offline key, done." And in 2015, that's exactly how it worked.

But today, that mental model is completely outdated.

The container signing architecture in the Docker world has gone through a turbulent decade: "They tried to do it the TUF way, developers refused to play along, the whole thing imploded, and the industry pivoted to a totally different approach." And that "different approach" turned out to be two competing approaches released around the same time, both fighting for dominance. Trying to keep up with this is exhausting.


Background: What "Signing a Container Image" Actually Means

Before diving into history, we need to nail down what "signing a container image" actually does. If this is fuzzy, the rest of the story will be too.

Structure of a Container Image

A container image is not just a tar file. A JSON file called the Manifest holds the hashes (digests) of each layer (filesystem diff) and config file that make up the image.

┌───────────────────────────────────────┐
│  Image Manifest (JSON)                │
│                                       │
│   config:  sha256:abc123...           │
│   layers:                             │
│     - sha256:def456... (base OS)      │
│     - sha256:789ghi... (app code)     │
│                                       │
│  ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─│
│  Manifest's own Digest:               │
│    sha256:xxxxxx...                   │
│  → This is the "image fingerprint"    │
└───────────────────────────────────────┘

Enter fullscreen mode Exit fullscreen mode

If even 1 bit of the image content changes, the Manifest's Digest changes completely. If we can guarantee just this Digest is correct, we can detect any tampering of the entire image.

"Signing" = Detached signature on the Digest

The intuitive idea is "embed the signature data inside the image," but that's impossible. If you change the image to insert a signature, the Digest changes, and the signature becomes invalid. Chicken-and-egg problem.

So container signatures are always Detached Signatures. Sign the Manifest's Digest from outside, and store the signature somewhere separate from the image itself.

So where is "somewhere separate"? This is the question that has been violently re-litigated for ten years.


Timeline: A Decade of Container Signing

Let's lay out the full picture first. Each entry will be expanded in later sections.

Year Event
2015.08 Docker Content Trust (DCT) released with Docker Engine 1.8. Notary v1, running underneath, is a pure TUF implementation. Signatures stored on a separate Notary server, not the registry.
2017.10 CNCF accepts Notary and TUF as Incubating projects.
2019.11 Notary v2 discussions kick off at KubeCon NA (San Diego). The following month, a kickoff meeting is held at Amazon's Seattle office with Docker, Microsoft, Amazon, Google, Red Hat, etc.
2021.06 Sigstore holds its first Root Key Ceremony (6/18). TUF is used only for "distributing root certificates."
2023.08 Notary v2 (Notation) v1.0.0 released (8/15). TUF completely dropped. Same month, Harbor 2.9.0 fully removes Notary v1 (deprecation began in 2.6.0).
2024.02 OCI Image/Distribution Specification v1.1.0 officially released. Referrers API standardized, formalizing in-registry signature storage.
2025.03 Azure Container Registry begins DCT deprecation (full removal scheduled for 2028.03).
2025.08 Docker Official Images' DCT signing certificate expires (8/8). DOCKER_CONTENT_TRUST=1 pulls start failing. DCT is effectively dead. Usage was less than 0.05% of all pulls.

Generation One: Notary v1 (Going All-In on TUF, 2015〜2025)

Architecture: A TUF Server "Next To" the Registry

In August 2015, Docker released Docker Content Trust (DCT). Setting DOCKER_CONTENT_TRUST=1 makes docker push automatically sign images and docker pull automatically verify them.

Underneath was Notary v1. It was a textbook TUF implementation: a Notary server running at a separate URL from Docker Hub, holding the full set of TUF metadata. Quick recap of the four roles:

Role File Purpose Key location
🏛️ Root root.json Anchor of trust. Declares public keys for the other 3 roles. Offline (in a vault)
🎯 Targets targets.json Records and signs the digests of images you want to protect. Offline
📸 Snapshot snapshot.json Guarantees consistency across metadata (prevents mix-and-match). Online
⏱️ Timestamp timestamp.json Freshness guarantee (prevents replay). Short expiration. Online

An "offline key" is a key kept on an air-gapped machine or in a physical vault; an "online key" is one that lives on a server for automated updates. Keeping the Targets key offline is the foundation of TUF's security model. This is exactly where things later explode.

Push Flow

push flow

The CLI computes the Manifest's Digest, signs an updated targets.json with the local Targets key, and uploads it. Step ② is an internal "use the key" operation (dotted line), not a network transfer.

Pull Verification Flow

pull verification flow

Walk Root → Timestamp → Snapshot → Targets, then compare the actual image's Digest from the registry against the record in targets.json. All four TUF roles in full motion: a spec-faithful architecture.

Why It Imploded

A theoretically correct architecture collapsed completely in practice.

1. It forced developers to manage signing keys

Every docker push prompted for the local Targets key passphrase. Maybe tolerable for solo developers, but for the modern "automate the push from CI/CD" workflow, this was fatal.

To wire it into CI, you had to put the Targets key (which was supposed to live offline in a vault) into the CI's secret store. "Putting the offline key online": a contradiction. This breaks the foundation of TUF's security model.

2. Lose the key = repository death

If you lose the Targets key, you can never sign images for that repository again. Key rotation must follow the TUF spec exactly, and the handoff overhead in team development was a nightmare.

3. Mismatch with the reality of distributed registries

This was the deepest structural problem. Container images don't deploy only to Docker Hub. AWS ECR, GCP Artifact Registry, Azure Container Registry, GitHub Container Registry, internal Harbor instances... registries are scattered everywhere.

In the Notary v1 model, every registry needed its own Notary server. Copy an image between registries, and the signature doesn't follow. The industry looked at that operational cost and said "no."

The Death of DCT: The Numbers Tell the Story

In the end, fewer than 0.05% of all Docker Hub pulls had DCT enabled.

On August 8, 2025, the oldest DCT signing certificates for Docker Official Images (nginx, ubuntu, etc.) expired. Users with DOCKER_CONTENT_TRUST=1 could no longer pull even the official images. Docker's response: "Please disable the DOCKER_CONTENT_TRUST environment variable." DCT quietly died.

Azure Container Registry began DCT deprecation in March 2025, with full removal scheduled for March 2028. Harbor moved earlier, fully removing Notary v1 in v2.9.0 back in 2023.


Generation Two: The OCI Registry-Native Era (2023〜present)

The Pivot: Put Signatures "Inside the Registry"

What the industry learned from Notary v1's failure: "Standing up a separate server just for signing doesn't work operationally."

The answer: store signature data directly in the same OCI registry as the image, as another OCI artifact (a blob conforming to the OCI spec). No extra registry to run. Copy the image between registries, and the signature comes along.

The OCI Distribution Specification v1.1.0, released in February 2024, formally standardized this approach. It introduced the Referrers API (GET /v2/<name>/referrers/<digest>), letting clients list all related artifacts (signatures, SBOMs, vulnerability scan results) attached to a given image's Digest.

Referrers API

Each artifact (signature, SBOM, etc.) points back to the parent image via a subject field. Verifier tools call the Referrers API to enumerate them and pick what they need to verify. No separate Notary server required.

Note: in production you usually pick either Cosign or Notation, not both (drawing them side-by-side just shows that both ride on the same spec). On top of this foundation, two signing projects are now competing for dominance.

Sigstore (cosign): The "Selective Bite" of TUF

Sigstore made a clean call. Stop using TUF's targets.json to manage image hashes. But don't throw TUF away entirely.

Sigstore uses TUF in exactly one place: safely distributing the root certificate of Fulcio (the signing CA) and the public key of Rekor (the transparency log) to clients. The first time you run cosign, a TUF client behind the scenes walks root.jsontimestamp.jsonsnapshot.jsontargets.json to fetch the certificates and public keys you should trust.

The heavy use case TUF was originally built for, "managing hashes of hundreds of thousands of packages," was abandoned. Sigstore kept only the lightweight role TUF excels at: "safely distributing root certificates."

Sigstore also gave a fundamental answer to the "key management is unbearable" problem that killed Notary v1: don't make developers hold private keys at all (keyless signing).

You authenticate via an OIDC (OpenID Connect, the standard protocol for ID token issuance) provider (GitHub, Google, etc.), Fulcio issues a short-lived certificate that expires in 10 minutes, and you sign with that certificate. The fact of signing is permanently recorded in Rekor's transparency log. The private key exists for a few seconds in memory and disappears. There is no key to manage in the first place.

sigstore

The revolutionary move: abandon the very idea of "protect the key" and replace it with "sign with a short-lived key, and leave only the signing trace in a public log forever."

Notary v2 (Notation): Total Abandonment of TUF

The next-generation Notary project, led by Docker and Microsoft. v1.0.0 released in August 2023. Active development continues as a CNCF Incubating project.

Notary v2 completely dropped the TUF specification. The four-role structure of Root, Targets, Snapshot, Timestamp is not used at all. Instead, it builds trust on X.509 certificate chains (the same mechanism as HTTPS certificates: trust propagates hierarchically from CA to intermediate CA to leaf certificate), a mechanism battle-tested for decades on the Web.

The mechanics are identical to SSL/TLS certificate verification. Signers hold X.509 certificates issued by a Certificate Authority (CA). Verifiers maintain a trust store (a list of CAs they trust) and walk the certificate chain attached to the signature to decide whether to trust it. TUF's complex chain of metadata is replaced with existing PKI infrastructure.

notation

You don't need to hold keys locally. Plugins connect to cloud KMS services like AWS Signer, Azure Key Vault, or HashiCorp Vault, delegating the signing operation. It also integrates with Kubernetes admission controllers (Ratify, Kyverno) so signature verification can be wired into deployment gates.

Comparing the Three Approaches

Notary v1 (DCT) Sigstore (cosign) Notary v2 (Notation)
Use of TUF Full implementation (4 roles) Root certificate distribution only Not used
Signature storage Notary server (separate infra) Inside OCI registry Inside OCI registry
Key management Developer manages locally None (keyless signing) Delegated to cloud KMS
Trust model TUF Root of Trust TUF + transparency log (Rekor) X.509 certificate chain
CI/CD fit ❌ Requires passphrase entry ✅ Fully automated via OIDC ✅ Automated via KMS plugin
Status (2026) ❌ Archived ✅ Adopted by npm, PyPI, Maven ✅ CNCF Incubating

Sidebar: Why Does TUF Work for PyPI?

If you've read this far, this question should be nagging you.

Notary v1 imploded over "key management is too hard." So how does Python's PyPI, which hosts over 500,000 packages, manage to make TUF (PEP 458) actually work?

The answer comes down to two structural differences.

1. Developers don't sign anything

PyPI's TUF deployment (PEP 458) is designed to protect the channel between the PyPI servers and the pip command. Developers just upload packages to PyPI as before. PyPI's backend automatically computes hashes and signs targets.json using PyPI's own online keys.

Developers don't even need to know TUF exists. This is the polar opposite of Notary v1, which forced developers to hold TUF's offline keys.

2. Centralized vs. distributed

Python packages all converge on a single central server: pypi.org. Run one TUF server, and you cover all 500,000 packages.

Container image registries, by contrast, are distributed across many places: Docker Hub, ECR, GCR, ACR, Harbor... Notary v1 required a TUF server per registry, and operational costs exploded.

PyPI Container ecosystem
Registry One: pypi.org Docker Hub, ECR, GCR, ACR, Harbor...
Who manages TUF PyPI server (automatic) Developers themselves (Notary v1)
Result ✅ Developers don't see TUF ❌ Developers burned out on key management

PyPI also spent years exploring "end-to-end signing by developers themselves" as PEP 480. But ultimately it gave up on forcing TUF-based key management onto developers and pivoted to Trusted Publishers (launched April 2023) using GitHub Actions OIDC. This is the same "OIDC + short-lived tokens" approach as Sigstore.

Docker, PyPI, npm: they all converged on the same conclusion. "Making developers manage private keys does not work." Notary v1's death is a lesson the entire industry has internalized.


Conclusion

"How do you protect the hash of a container image with TUF's Targets?"

In the old days, you protected it with targets.json (Notary v1). But in a distributed container ecosystem, the model that asks developers to manage offline keys completely fell apart. Today, instead of managing the image Digest directly with TUF, signatures are stored directly in the OCI registry (Sigstore / Notary v2).

Security that nobody uses is not security. The decade of Notary v1 proved that.


References