2nd/10_Wiki/Topics/AI_and_ML/공급망 공격 (Supply Chain Attack).md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-공급망-공격-supply-chain-attack	공급망 공격 (Supply Chain Attack)	10_Wiki/Topics	verified	self	Supply Chain Attack SCA 의존성 공격 dependency confusion	none	A	0.92	applied	security supply-chain devsecops sbom		2026-05-10	pending	language framework python sigstore

공급망 공격 (Supply Chain Attack)

매 한 줄

"매 빌드 파이프라인의 매 한 곳이 매 약점이다". 공격자는 직접 target 을 뚫는 대신 매 dependency, 매 build agent, 매 registry 를 오염시켜 매 downstream 수천 개 product 에 한 번에 침투. SolarWinds(2020) → xz-utils(2024) → npm event-stream / ua-parser-js / Polyfill.io 사슬을 거치며 매 SBOM·sigstore·SLSA L3+ 가 매 2026 표준이 되었다.

매 핵심

매 공격 표면

• Source: 매 maintainer 계정 탈취, malicious commit (xz-utils Jia Tan).
• Build: 매 CI runner 침투 (CodeCov bash uploader, GitHub Actions token 유출).
• Package: 매 typosquatting (reqeusts), 매 dependency confusion (private name 을 public registry 에 선점).
• Distribution: 매 mirror / CDN 변조 (Polyfill.io 2024).
• Update channel: 매 auto-update 서버 hijack (SolarWinds Orion).

매 1차 방어

• SBOM (CycloneDX / SPDX) — 매 component 추적, EU CRA 2026 mandate.
• Sigstore cosign — 매 keyless signing, transparency log (Rekor).
• SLSA L3+ — 매 hermetic, isolated, provenance-attested build.
• Pinning + lockfile — 매 hash-pin (pip --require-hashes, npm ci).

매 응용

1. Open-source 의존성 audit pipeline.
2. 내부 enterprise artifact registry hardening.
3. ML model supply chain (huggingface, model card 위조 방어).

💻 패턴

sigstore cosign 으로 컨테이너 image sign + verify

# Sign (keyless OIDC)
cosign sign --yes ghcr.io/org/app@sha256:abc123

# Verify in admission controller
cosign verify ghcr.io/org/app@sha256:abc123 \
  --certificate-identity-regexp '^https://github\.com/org/' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

SBOM 생성 + 취약점 scan (syft + grype)

syft packages dir:. -o cyclonedx-json > sbom.json
grype sbom:sbom.json --fail-on high

Dependency confusion 방어 (npm scoped + .npmrc)

# .npmrc — 매 internal scope 만 private registry 사용
@acme:registry=https://npm.acme.internal
//npm.acme.internal/:_authToken=${NPM_TOKEN}
registry=https://registry.npmjs.org/

Python hash-pinned install

pip-compile --generate-hashes requirements.in
pip install --require-hashes -r requirements.txt

GitHub Actions OIDC + 최소 권한

permissions:
  contents: read
  id-token: write   # OIDC 만, GITHUB_TOKEN 권한 격리
jobs:
  build:
    steps:
      - uses: actions/checkout@v4
        with: { persist-credentials: false }
      - uses: aws-actions/configure-aws-credentials@v4
        with:
          role-to-assume: arn:aws:iam::123:role/ci-deploy
          aws-region: us-east-1

SLSA provenance (in-toto attestation)

import json, hashlib
from in_toto_attestation.v1 import statement_pb2 as s

stmt = s.Statement(
    type="https://in-toto.io/Statement/v1",
    subject=[s.ResourceDescriptor(name="app",
        digest={"sha256": hashlib.sha256(open("app","rb").read()).hexdigest()})],
    predicate_type="https://slsa.dev/provenance/v1",
    predicate={"buildDefinition": {"buildType": "github-actions-v1"}},
)

Maintainer takeover 탐지 (commit signature drift)

def detect_anomaly(commits):
    # 매 갑자기 unsigned commit, 매 새로운 GPG key, 매 timezone 급변
    keys = {c.gpg_key for c in commits if c.gpg_key}
    if len(keys) > 3 or any(c.gpg_key is None for c in commits[-10:]):
        alert("Maintainer key drift")

매 결정 기준

상황	Approach
OSS 의존성 多	SBOM + grype CI gate
내부 private package	scoped registry + dependency confusion 차단
Container 배포	cosign keyless + admission verify
규제 산업 (gov, finance)	SLSA L3+ hermetic build, reproducible
ML model 배포	model signing + dataset provenance

기본값: SBOM(syft) + cosign keyless + lockfile hash-pin + OIDC short-lived credential.

🔗 Graph

• 부모: CI/CD Pipeline & IDE Security Integration
• 변형: Dependency Confusion
• 응용: SBOM · Sigstore · SLSA
• Adjacent: Zero Trust

🤖 LLM 활용

언제: SBOM 차이 분석, CVE → affected component mapping, supply chain risk 자동 triage. 언제 X: 매 cryptographic signature 검증 자체는 매 deterministic tool (cosign) 의 사용 — LLM 추론 X.

❌ 안티패턴

• Latest tag 사용: image:latest — 매 mutable, 매 unverifiable. Pin digest.
• Curl | bash: 매 unsigned script 실행 — checksum 최소.
• Long-lived CI token: PAT 영구 보관 → OIDC short-lived 로 교체.
• 단일 maintainer OSS 채택 without audit: bus factor 1 = supply chain risk.

🧪 검증 / 중복

• Verified (CISA 2025 SCRM guidance, SLSA v1.0 spec, NIST SSDF).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — full content (SBOM, sigstore, SLSA, dependency confusion patterns)