Welcome to the July edition of the Xygeni Malicious Code Digest. This month, our security research team confirmed over 780 malicious packages across npm, PyPI, and OpenVSX (the VS Code extension marketplace), tracked across five weekly digests.
July was defined by three converging trends: sustained, high-volume version-flooding campaigns designed to outlast takedowns; a sharp escalation in attacks targeting AI tooling, MCP servers, and agentic workflows; and coordinated dependency-confusion campaigns against both enterprise namespaces and crypto/DeFi ecosystems.
Ma waena o nā hoʻolaha kaulana loa i hoʻopaʻa ʻia i kēia mahina:
bingo-aion PyPI resurfaced twice, flooding the registry with well over 150 versions across two bursts (July 13 and July 21), confirming this is an ongoing operation, not a one-off.zevairouterbecame July’s largest single-package campaign: over 65 versions across npm, published continuously from July 25–28.gcli-control, the Windows RAT we profiled in detail that routes its C2 through npoint.io, escalated from version 0.1.0 to 0.13.0 across three separate weeks.@szc-ft/mcp-szcd-client, the package behind SkillLeak, our writeup on a credential decryptor delivered through a bundled MCP skill, was confirmed July 2.- The week of July 7 brought the month’s heaviest AI-tooling targeting:
mcp-server-pg,anthropic-toolkit,openai-agents-helpers,ollama-helpers, a@langgraphjs/toolkit, impersonating MCP, Anthropic, OpenAI, Ollama, and LangGraph. - A 17-package PayPal impersonation cluster hit npm July 27, all at version 28.0.0 within minutes.
@wagni_bot, roughly 60 npm packages impersonating crypto wallet SDKs (Ethereum, Solana, Binance, and more), all published in a single day, July 10.- Over a dozen fake VS Code extensions surfaced on OpenVSX, confirming attackers are expanding beyond package registries into the IDE itself.
The defining pattern of July: attacks increasingly target the AI agents and automated tooling that install packages with no human reviewer in the loop, at a publishing velocity built to outrun manual review.
Below is a summary of what we found. You can see all five weeks’ data disclosed in full detail at the Malicious Code Digest index.
Pule 5: Ua ʻike ʻia ma luna o 180 mau pūʻolo
Pule 4: Ua ʻike ʻia ma luna o 165 mau pūʻolo
Pule 3: Ua ʻike ʻia ma luna o 145 mau pūʻolo
Pule 2: Ua ʻike ʻia ma luna o 200 mau pūʻolo
Pule 1: Ua ʻike ʻia ma luna o 90 mau pūʻolo
From Version Storms to AI Impersonation: What July’s Supply Chain Attacks Reveal
The campaigns above aren’t edge cases, they’re the baseline now. Version-flooding storms, coordinated impersonation drops, and AI-tooling lookalikes are hitting real teams in real SDLCs every week, often with no human in the loop between publication and install.
ʻO Xygeni ka ʻike ʻana i nā polokalamu ʻino a supply chain security platform gives organizations the visibility to catch malicious dependencies before they execute on a developer machine, enter a build system, or reach production. Coverage spans npm, PyPI, OpenVSX, and beyond, monitoring for suspicious publishing patterns, namespace abuse, typosquatting, and AI-native attack techniques as they emerge.
Hoʻonohonoho pono ʻia kēlā me kēia ʻike e ka hiki ke hoʻohana ʻia, ka hiki ke hiki ke hoʻokokoke ʻia, a me ka hopena o ka ʻoihana, no laila e nānā ana kāu hui i nā mea e pono ai ke hoʻoponopono, ʻaʻole i ka walaʻau.
E ʻimi i kēlā me kēia pūʻolo ʻino a me ka hoʻolaha i hōʻoia ʻia e ka Xygeni Security Team ma ka Hōʻuluʻulu Manaʻo Koda ʻIno.
E noho palekana. E noho wikiwiki. E noho mana me Xygeni.
