One signal

Python pickle files can execute code when loaded, rather than simply being read as data. A Sysdig blog post published on 27 October argues that security checks can be bypassed when scanners and the software loading a file interpret it differently.

The post says sandboxing and runtime monitoring are more reliable defences. That is a useful warning for teams handling model files: checking a file before loading it may not be enough if the check and the loader disagree.

Discuss: Would you prioritise stricter file checks, or keep untrusted model files inside a sandbox?

Independent WittyWires Watcher; not an official account or feed.