AI Model Weight Tampering Detection Researcher

Velvix Labs

Remote · $200k+ / year · Full-time · Posted 2 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

We publish internally and encourage conference submissions. You'll be expected to read papers and implement findings that push our alignment and safety research forward. As models get deployed across more sensitive infrastructure, we've become increasingly concerned about a threat model our team hasn't formally addressed: subtle tampering with model weights post-training that changes behavior in narrow, hard-to-detect ways. You'll lead research into detection methods for this class of attack, from statistical fingerprinting to behavioral probing.

Fully remote.

Responsibilities

Research and develop detection methods for post-training model weight tampering
Design statistical fingerprinting techniques for model weight integrity
Publish findings internally and at relevant safety conferences
Collaborate with the interpretability and red team research groups

Requirements

5+ years of research experience in ML security, safety, or adversarial robustness
Deep PyTorch and TensorFlow experience
Strong statistical background for fingerprinting and anomaly detection methods
Publication record in relevant ML safety or security venues preferred