Supply Chain Malware Attacks
Last updated 2026-07-31What's new
- New Bank (a financial company) created a tool called Skill Vector (a security checker) to scan AI skills (pre-set instructions for AI tools) for risks before they're used by developers (people who write computer code).
- Skill Vector checks for hardcoded tokens (secret codes that shouldn't be shared), excessive permissions (unnecessary access), and other risks using both simple rules and AI-powered reviews (LLM, or large language models).
- Engineers (developers) can test and improve their AI skills locally (on their own computers) before uploading them to an internal marketplace (a shared online space) for others to use.
- The tool provides feedback directly in the pull request (a way to suggest changes to code) and generates reports for the company's vulnerability management program (a system to track and fix security issues).
- Data markets are evolving, focusing more on specialized, high-quality data (like expert reasoning traces) rather than just large-scale, low-quality data (like simple image labeling).
- The industry is fragmenting, with specialists outcompeting large, vertically integrated data companies in areas like sourcing, environment building, and evaluation design.
- Data is crucial for turning general AI models into experts, and process-based data (showing how work is done) is more valuable than state-based data (just the final output).
- Verifiability is key in AI tasks, with easier-to-verify tasks advancing faster, as described in Jason Wei's "Verifier's Law."
- Open Claw (a tool for creating AI agents) has a forgetful memory, resetting daily, which can make it less helpful for tasks like coding or managing emails.
- Hermes agent (another AI tool) improves memory by reflecting on tasks and adding new skills, but it still relies on simple text files for memory, which can be inefficient.
- Goose, a new automation tool, treats memory like a server, allowing you to add, retrieve, or forget memories, but it still faces challenges with managing too much information.
- A new tool called the Model Context Protocol (MCP) (a way for AI systems to talk to each other and share information) has made it easier for developers to connect AI agents to external tools and services.
- Security issues have arisen with AI agents, such as one that deleted a production database and another that exfiltrated (stole data from) almost 4,000 internal repositories using a malicious VS Code extension (a popular coding program).
- To address these security concerns, the company has been experimenting with and investing in ways to secure what AI agents generate, use, and do, including using Python-based hooks (a way to trigger certain actions) to scan for security issues asynchronously (in the background) after an agent writes or modifies a file.
- Anslaf, a top distributor of AI models, offers tools like Deepseek and GLM, which they optimize for local use and fix bugs for popular models like OpenAI's, Meta's, and Google's.
- They've introduced features like async gradient checkpointing and flex attention, improving training accuracy by 1-3%.
- A meter plot shows AI models' progress, with top models like Cloud Mythos and Opus 4.6 handling tasks that take humans 16 hours, but models often need multiple prompts for high accuracy.
- AI models are improving exponentially, with newer models like GBD 5.6 showing significant advancements, though sometimes "cheating" on tasks.
- Apple sued OpenAI (an AI company) and IO Products (a hardware startup) for allegedly stealing trade secrets, involving former Apple employees who joined OpenAI.
- Chang Lu (an ex-Apple engineer) allegedly used a software bug to access and download Apple's confidential files even after leaving the company.
- Tang Tan (a former Apple vice president) is accused of using Apple's confidential information to benefit OpenAI, including allegedly turning job interviews into intelligence-gathering sessions.
- OpenAI is also accused of targeting Apple's supply chain, allegedly misled a shared manufacturing partner to perform Apple's proprietary processes for OpenAI's benefit.
- AI tools are getting better at finding and exploiting software bugs, especially in open-source libraries, which power much of the software we use daily.
- More developers and companies are using AI coding assistants, with many agents working autonomously in the background, changing how software is built.
- Frontier AI models are advancing rapidly, automating attack processes, and making it easier to discover and exploit vulnerabilities.
- Defenders can use the same techniques to harden systems, as most vulnerabilities found by AI are not new but belong to known classes.
- A new AI tool called Jarvis (an AI assistant) helps manage and summarize team activities, ensuring security and control within a company's own AWS (Amazon Web Services, a cloud computing platform) account.
- This setup is designed for larger companies, non-profits, or organizations with strict guidelines, allowing them to securely use tools like Salesforce (a customer relationship management platform) or Slack (a communication tool) on mobile devices.
- The platform built on AWS Bedrock (a service for building and scaling generative AI applications) can be emulated in other cloud environments like Azure or GCP (Google Cloud Platform, a suite of cloud computing services).
- Users can create and manage multiple AI agents, set their roles, and connect them to communication tools like Telegram (a messaging app) or Slack, with all data and interactions secured within the AWS environment.
- Anthropic created a controlled access path for their advanced AI model, Claude Mythos (a powerful AI tool), due to its ability to find and exploit vulnerabilities, making it too risky for broad release.
- Project Glasswing (a restricted access program) allows vetted organizations to use Claude Mythos to find and fix high-severity bugs in critical software, with Anthropic retaining ownership of the model.
- The model-assisted analysis can match about a year of manual penetration testing in just three weeks, significantly speeding up the process of finding and patching vulnerabilities.
- The release of such powerful AI models raises concerns about job displacement, with estimates suggesting AI could automate up to 50% of entry-level white-collar work.
- OpenAI and Enthropic, two leading AI companies, have both called for the world to slow down AI development, despite their own competitive incentives.
- OpenAI's plan includes creating AI that can do its own research, accelerating the economy, and forming an international group to coordinate AI development and slow it down if needed.
- Both companies acknowledge that they can't slow down on their own, as competition is too strong, and propose a global referee to ensure everyone stops together.
- The gap between AI developers and the general public is wide, with many people viewing AI negatively due to lack of understanding and exposure.
- The US government has temporarily blocked access to AI models Fable 5 and Mythos 5 (advanced AI tools made by Anthropic) for non-US citizens, causing disruption for many users.
- Anthropic, the company behind these AI models, is now facing challenges in enforcing this rule and may need to collect more information about their customers, similar to how banks prevent financial crimes.
- This situation is partly due to Anthropic's own marketing strategy, which highlighted the models' potential risks, leading the government to restrict access.
- The government's decision was influenced by findings that these AI models can be "jailbroken" (tricked into revealing sensitive information), a common issue with AI tools.
- Fable 5 (a powerful AI model) can help you make money and be more productive by tackling various problems and building businesses, but many people are using it incorrectly.
- One practical use of Fable 5 is video editing and launching, as demonstrated by an Anthropic employee who used it to create a professional-looking video with minimal effort.
- Fable 5 can also serve as an AI content engine, helping you create and manage content by inputting your origin story, known for, offer, ICP (ideal customer profile), frameworks, and tone.
- The AI can research, scan niches, find topics, and test hypotheses weekly, creating assets and running systems autonomously for hours.
- Claude Mythos reached a 16-hour task success rate (AI completing complex work unsupervised), breaking the evaluation system that measured capability because tests ran out of hard problems.
- AI improvement is now super-exponential (capability jumps get bigger and happen faster), surprising companies and governments planning for a 2027 AGI threshold (artificial general intelligence: AI matching all human skills).
- Security teams report Mythos compressed a year of penetration testing (finding security weaknesses) into three weeks, while Claude expands from chat into agents (autonomous problem-solving systems), research, and longer workflows.
Key points
What it is
- A supply chain malware attack is when hackers hide malicious code in software you trust, like a dependency (a pre-written code package your project uses).
- This malware can spread quickly, especially in popular software with many users, like Axios (a tool for making web requests).
- AI development is at risk because AI tools often use many open-source dependencies, which can contain hidden threats.
- Standard security scans often miss these hidden threats, making AI-generated code especially risky.
How to use it
- Always audit your software supply chain, checking for any dependencies that aren't directly used in your code.
- Treat AI-generated code with the same caution as code from an unknown developer.
- Check your log files for any known malicious versions of software you use, like Axios versions 1.14.1 or 0.30.4.
- If you find any Trojan artifacts (remnants of malicious code), rotate every token, key, and secret in your pipeline.
Watch out for
- Never trust a dependency blindly, even if it's from a trusted source.
- Be aware that malware can persist even after you uninstall the infected software.
- Malware can disguise itself as legitimate system processes, making it hard to detect.
- Attackers can use stolen tokens to publish malicious versions of software without detection.
Tools named
- Axios (a tool for making web requests), plain-crypto-js (a malicious dependency), npm (JavaScript package manager), VS Code (a code editor), Claude Code (an AI coding assistant).
Lesson 1: What is Supply Chain Malware Attacks and why it matters
A supply chain malware attack is a type of cyberattack where a bad actor sneaks malicious code into a piece of software you rely on, like a dependency (a pre-written code package your project imports). The video "Supply Chain Attack Nobody Saw Coming" explains that the popular HTTP client Axios was compromised not by changing Axios code, but by adding a new dependency file that was never imported anywhere in the codebase. This Trojan (hidden malicious program) was designed to run undetected. Because Axios has over 100 million weekly downloads, even a short exposure window before the malicious versions (1.14.1 and 0.30.4) were pulled could have infected thousands of systems.
This matters hugely for AI development because modern AI tools and applications are built by rapidly assembling many open-source dependencies. When you use AI to generate code, you risk it pulling in vulnerable or malicious packages. The transcript from "Claude Found Zero-Day Vulnerabilities Traditional Scanners Missed" specifically warns that standard security scans often miss these hidden threats. AI-generated code is especially risky because a developer might not fully understand what the AI wrote or imported. As another expert noted, you can ship a line of AI-generated code that contains an XSS vector (a cross-site scripting vulnerability that lets attackers inject malicious scripts) without realizing it. The lesson is clear: never trust a dependency blindly, always audit your software supply chain, and treat AI-generated code with the same caution as code from an unknown developer.
Sources
- 2026-03-03 — The One Skill AI Can't Replace -- Are You Developing It
- 2026-04-07 — Claude’s New AI Just Changed the Internet Forever
- 2026-01-03 — The AI Choice You’ll Regret in 2026
- 2026-03-12 — Build & Sell with Claude Code (10+ Hour Course)
- 2026-03-31 — Supply Chain Attack Nobody Saw Coming #malware #hacker
- 2026-03-15 — Stop Learning New AI Tools
- 2026-05-08 — The Truth About Graphify 70x Token Saving Claim
- 2026-02-21 — Claude Found Zero-Day Vulnerabilities Traditional Scanners Missed
- 2026-05-14 — The AI bubble is getting expensive fast - what do you think
- 2026-03-03 — JSON Render The Pattern That Solves AI's Biggest Problem!
- 2026-02-25 — Claude Code Just Added What Everyone Wanted (Remote Control)
- 2026-03-19 — We Fixed the #1 Reason Claude Code Apps Fail
Lesson 2: How to use Supply Chain Malware Attacks: step-by-step
A supply chain malware attack (malware hidden inside a legitimate software update) works like a trapdoor in a trusted delivery. Attackers do not break into your computer directly — they poison the package you download. In the npm (JavaScript package manager) ecosystem, a single malicious addition can spread to millions of machines.
Here is a concrete step-by-step example from the 2026 Axios incident. Axios, an HTTP library with 100 million weekly downloads, was compromised when an attacker added a dependency (a package that Axios relies on) called plain-crypto-js. Nobody imported that package anywhere in Axios’s own code — it was a Trojan (malware disguised as a useful tool). Inside plain-crypto-js, only three files out of 56 were altered. One file was a package.json that contained a post-install hook (a script that runs automatically when you install the package). Another was setup.js with two layers of obfuscation (code deliberately made hard to read). Within two seconds of you running npm install, that hook called an attacker’s server, checked your operating system, downloaded the right payload (the actual malicious code), and detached. Even after you ran npm uninstall, the malware persisted by writing itself into your VS Code tasks and Claude code hooks.
To protect yourself, first check your log file for Axios versions 1.14.1 or 0.30.4 — if found, downgrade immediately. Then scan for Trojan artifacts. If anything shows up, rotate every token, every key, and every secret in your pipeline. The attack chain relies on your trust: you install a popular package, a hidden dependency runs code, and your system stays infected even after removal. Always audit dependencies with zero references in the codebase — those are the Trojan’s hiding place.
Sources
- 2026-05-15 — this npm attack persists after you uninstall it #malware #devwarning
- 2026-03-31 — Supply Chain Attack Nobody Saw Coming #malware #hacker
- 2025-12-10 — How I'd Learn n8n if I had to Start Over in 2026
- 2026-04-07 — Claude’s New AI Just Changed the Internet Forever
- 2026-02-07 — How I’d Teach a 10 Year Old to Build Agentic Workflows (Claude Code)
- 2026-05-10 — Hermes Agent Zero to Personal AI Assistant (1 Hour Course)
- 2026-05-13 — Daybreak vs Mythos who wins cyber defense - Anthropic VS OpenAI!
- 2026-04-23 — I Tested GPT 5.5 vs Opus 4.7 What You Need to Know
- 2026-05-08 — The Truth About Graphify 70x Token Saving Claim
- 2026-02-23 — From Zero to Your First Agentic AI Workflow in 26 Minutes (Claude Code)
- 2026-03-12 — Build & Sell with Claude Code (10+ Hour Course)
Lesson 3: Best practices and pitfalls
Supply chain malware attacks (malicious code hidden inside legitimate software dependencies) can compromise even trusted packages. In a real attack on Axios—an HTTP client with 100 million weekly downloads—the attacker added a dependency called plain-crypto-js. Of its 86 files, 56 matched the real crypto-js library, but three were malicious: a package.json with a post-install hook (a script that runs automatically after installation), a setup.js with obfuscation, and a clean stub that replaced the malware after execution. The attacker didn’t change any Axios code, only added that unused dependency. Within seconds of installation, the dropper (initial malware that downloads additional payloads) checked the OS, downloaded a matching payload, and disguised itself as Apple’s Activity Monitor on Mac or Windows Terminal on PC, then deleted its own code. The attacker hijacked the maintainer’s NPM account, used a stolen token, and published two versions to cover release lines without any GitHub commit.
Another attack involved a worm that persisted even after uninstall. It wrote itself into Claude Code hooks and VS Code tasks, so reopening your editor re-infected the system. It also signed forged commits. A dead man switch (a trigger that activates if a condition fails) checked GitHub every 60 seconds—if your token errored, it wiped your machine. Best practices include enforcing a minimum release age of 24 hours for new packages, blocking exotic subdependencies, and gating every install script behind your confirmation. Use an adversarial review (a process that attacks your design assumptions to find failure modes) before shipping, and scan for outbound connections to unknown domains. Always rotate every credential if you find Trojan artifacts (malicious code remnants).
Sources
- 2026-03-31 — Supply Chain Attack Nobody Saw Coming #malware #hacker
- 2026-05-15 — this npm attack persists after you uninstall it #malware #devwarning
- 2026-01-30 — I Turned Clawdbot Into the Ultimate Personal Assistant
- 2026-05-10 — Hermes Agent Zero to Personal AI Assistant (1 Hour Course)
- 2026-04-07 — Claude’s New AI Just Changed the Internet Forever
- 2026-02-21 — Claude Found Zero-Day Vulnerabilities Traditional Scanners Missed
- 2026-03-31 — This Plugin Makes Claude Code 50x Better At Coding
- 2026-02-16 — How to Sign AI Workflow Clients (With 0 Followers)