Research and updates from the team
Papers, findings, and news from the people building Sentinel.
A general theory of data corruption, published in JMLR
Our paper building a general theory of data corruption in supervised learning was published in the Journal of Machine Learning Research. Here's what it found.
ResearchGateBreaker: our attack on Mixture-of-Experts LLM safety, accepted at USENIX Security 2026
Our paper on attacking safety alignment in Mixture-of-Experts LLMs was accepted at USENIX Security 2026. Here's what GateBreaker found.
ResearchNeuroStrike: our attack on LLM safety alignment is heading to NDSS 2026
Our paper on neuron-level attacks against aligned LLMs was accepted at NDSS Symposium 2026. Here's what it found.
ResearchBest paper at COPA 2024: conformal prediction under data contamination
Our award-winning paper on conformal prediction under contaminated calibration data. Here's what it found.
ResearchTesting generative models that don't give you a probability, published at NeurIPS 2022
Our NeurIPS 2022 paper on testing generative models that only give you samples, not probabilities. Here's what it found.
ResearchOur book on learning without clean labels, published by MIT Press
Our book on learning from incomplete and imperfect labels was published by MIT Press. Here's what it covers.
ResearchTesting whether a model fits data on curved spaces, published at ICML 2021
Our ICML 2021 paper on goodness-of-fit testing for models on curved geometric spaces. Here's what it found.
ResearchBest paper at NeurIPS 2017: a goodness-of-fit test that scales linearly
Our NeurIPS 2017 best paper on a goodness-of-fit test that scales linearly with sample size. Here's what it found.