security deserialization ai_generated true

Remote code execution — unsafe deserialization of untrusted data

ID: security/deserialization-rce-untrusted-input

Also available as: JSON · Markdown

88%Fix Rate

92%Confidence

4Evidence

2023-01-01First Seen

Version Compatibility

Version	Status	Introduced	Deprecated	Notes
any	active	—	—	—

Using pickle or Java ObjectInputStream on untrusted input allows arbitrary code execution.

generic

95% success Replace unsafe deserializers with safe formats like JSON or Protobuf
```
json.loads(data) instead of pickle.loads(data)
```
Sources: https://docs.python.org/3/library/pickle.html#restricting-globals
78% success If pickle is required, use RestrictedUnpickler with strict class allowlist
```
class SafeUnpickler(pickle.Unpickler): def find_class(self, module, name): ...
```
Sources: https://docs.python.org/3/library/pickle.html#restricting-globals

Common approaches that don't work:

Validate the serialized data before deserializing 90% fail
The exploit triggers during deserialization itself, not after
Use a custom ObjectInputStream with class whitelist alone 65% fail
Gadget chains can use allowed classes to achieve code execution