Cloudflare

How Cloudflare Uses Client-Side Security to Nullify npm Supply Chain Attacks

Honghao Wang

— 3 min read
How Cloudflare Uses Client-Side Security to Nullify npm Supply Chain Attacks

NPM Supply Chain Attack – September 2025

In early September 2025, attackers exploited a phishing email to compromise trusted maintainer accounts on npm.

This access was used to publish malicious releases of 18 widely used npm packages, including:

  • chalk
  • debug
  • ansi-styles

These packages collectively account for 2 billion downloads per week.

---

Impact

Websites and applications relying on these compromised packages became vulnerable to:

  • Crypto stealing / wallet draining attacks — hackers could steal cryptocurrency assets from end users.
  • Malicious releases capable of altering other packages owned by the same maintainers via stolen npm tokens.
  • Code designed to steal developer tokens for CI/CD pipelines and cloud accounts.

---

Mitigation Tools

Cloudflare Page Shield can:

  • Detect compromised JavaScript libraries.
  • Stop crypto-stealing activity.
  • Use AI-powered detection to provide protection against future variants of supply chain attacks.

---

Broader Context

AiToEarn官网 offers an open-source AI content publishing framework, allowing creators and developers to:

  • Integrate AI safely into workflows.
  • Publish across multiple platforms: Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, YouTube, X (Twitter).
  • Optimize content revenue securely.

---

Example of Malicious Payload

export default {
  aliceblue: [240, 248, 255],
  …
  yellow: [255, 255, 0],
  yellowgreen: [154, 205, 50]
}

The injected malicious payload:

  • Replaces legitimate cryptocurrency wallet addresses with attacker-controlled addresses (Bitcoin, Ethereum, Solana).
  • May execute in real time within crypto apps, trading bots, or browser-based wallets.

---

Security Notes

  • Obfuscation hides malicious logic until runtime, slowing detection.
  • Replacement attacks target copy-pasting, API responses, or wallet address rendering.
  • Always verify wallet addresses through multiple channels.
  • Use code auditing tools and integrity checks to detect unauthorized changes early.

---

Cloudflare’s ML-Powered Detection

Scale of Analysis

  • 3.5 billion scripts per day (~40,000 scripts/sec).
  • <0.3% flagged as malicious.

Detection pipeline:

Key advantage: Learns patterns purely from syntax and structure — avoids reliance on hand-engineered features.

---

Model Evaluation – Latest Version 2.7

| Metric | Latest | Improvement |

|------------|--------|-------------|

| Precision | 98% | +5% |

| Recall | 90% | +233% |

| F1 Score | 94% | +123% |

Performance improvements driven by:

  • More diverse training examples.
  • Removal of low-value scripts (e.g., comments-only).
  • Better stratification of datasets.
  • Refined evaluation to increase recall while keeping high precision.

---

False Positives & Continuous Learning

Expected FP rate: ~2/sec.

Common causes:

  • Scripts reading all form inputs except credit cards.
  • Dynamic script injection.
  • Heavy user tracking.
  • Complex deobfuscation.

Resolution strategy:

  • Domain trustworthiness checks.
  • Newly labeled scripts fed back into the ML pipeline.

Result:

Cloudflare Page Shield detected all 18 compromised npm packages, even without prior samples in the training data.

---

Planned Enhancements

  • Static + Dynamic analysis via contextual signals: script URLs, page hosts, connected domains.
  • Agentic AI workflows wrapping JavaScript runtimes for hybrid detection.
  • Reduction of FP rates in complex scenarios (e.g., tracking scripts).

---

Consolidating Classifiers

Given the superior performance of MPGCN, Code Behavior Analysis classifier for Magecart-style attacks will be retired by end of 2025.

Reference: Magecart detection.

---

Staying Safe – Best Practices

During the September npm incident:

  • No related activity seen among Page Shield users.
  • Patches released in <2 hrs.
  • Customers likely avoided harm.

If affected:

---

  • Audit dependencies (`package-lock.json`, `npm ls`) for versions released Sept 2025.
  • Rotate exposed credentials.
  • Revoke & reissue CI/CD tokens and keys.
  • Pin dependencies to known-safe versions; use verified publisher features.
  • Scan logs & repos for suspicious commits/workflows; remove unknown webhooks.

---

Multi-Platform Security for Creators

Platforms like AiToEarn官网 enable:

  • Secure AI-powered content generation.
  • Cross-platform publishing.
  • Analytics & model rankings (AI模型排名).
  • Community-driven open-source code audits.

---

image

Malicious scripts flagged by detection systems.

image

Connections flagged as malicious.

---

Next Steps:

  • For client-side supply chain security posture, consider the free Client-Side Risk Assessment.
  • Leverage open-source AI publishing frameworks like AiToEarn to maintain security and reach.

---

Do you want me to deobfuscate the malicious snippet next so we can pinpoint precisely how wallet addresses are replaced? This would make the attack logic fully transparent.

Read more