Distribution-Specific Curvature Control with Finite-Sample Guarantees for Open-Weight Safety

July 24, 2026 ยท Grace Period ยท ๐Ÿ› AAAI 2027

โณ Grace Period
This paper is less than 90 days old. We give authors time to release their code before passing judgment.
Authors Domenic Rosati, Ali Dadsetan, Hong Huang, Xijie Zeng, Hassan Chowdhry, Subhabrata Majumdar, Hassan Sajjad, Frank Rudzicz arXiv ID 2607.22929 Category cs.LG: Machine Learning Citations 0 Venue AAAI 2027
Abstract
A short fine-tuning run can undo the safety guards of an open-weight model---retraining a refusal-trained assistant to aid weapons development or produce hate speech. Preventing such harmful fine-tuning while retaining benign adaptability remains difficult: the only prior method with an explicit curvature certificate, spectral deformation, inflates curvature globally and thereby obstructs benign adaptation along with harmful adaptation. We propose HarmAlign, which applies function-preserving spectral deformation along a estimated contrastive activation subspace. We derive finite-sample bounds for the estimated subspace energy and the resulting local harmful-distribution curvature lower bound. A stability--progress dichotomy for constant-step gradient descent turns the certified curvature into conditional convergence-rate control. Empirically, within a fixed-architecture, finite-budget first-order threat model, HarmAlign blocks direct fine-tuning and three data- or objective-adaptive attacks across a hazardous-knowledge relearning setting and a harmful-assistance fine-tuning setting, while the protected benign tasks remain trainable. The block persists across the tested first-order optimizer variants over every attack checkpoint, and under out-of-distribution harmful fine-tuning, and it extends to important cases in our threat model: accidental safety degradation and emergent misalignment.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

๐Ÿ“œ Similar Papers

In the same crypt โ€” Machine Learning