Detecting Backdoors with Meta-Models

"It is widely known that it is possible to implant backdoors into neural networks, by which an attacker can choose an input to produce a particular undesirable output (e.g. misclassify an image). We propose to use meta-models, neural networks that take another network's parameters as input, to detect backdoors directly from model weights. To this end we present a meta-model architecture and train it on a dataset of ~4000 clean and backdoored CNNs trained on CIFAR-10. Our approach is simple and scalable, and is able to detect the presence of a backdoor with accuracy"

https://openreview.net/forum?id=cmJiEqniEc

#infosec #ai

I was about to increase the inactivity timeout on my phone, and realized it was better to make my PIN shorter and more crackable.

If the device isn't locking on failed attempts then all is lost, and if it is locking then an extra-hard password will lead to insecure practices like a longer inactivity timeout.

Most of the value of the OpenSSF Best Practices Badge and Security Scorecard is influence over your project's roadmap. You use them to find, organize, and prioritize security improvements.

But neither of those projects thinks of itself that way.

What their UX *should* be is a Kanban board for evaluating each criteria, then doing project management to mitigate work items you find.