Self-Calibrating Weight-Arithmetic Safety-Drift Repair

Pratinav Seth; Vinay Kumar Sankarapu

← Publications / Safety Post-Training & Alignment

Self-Calibrating Weight-Arithmetic Safety-Drift Repair

Pratinav Seth, Vinay Kumar Sankarapu

Under Review2026My role: Lead authorUnder review

Self-calibrating repair: the aligned-minus-base task vector is masked and scaled by a 1-D search against an MMLU floor

Cite

@misc{seth2026selfcalibrating,
  title         = {{Self-Calibrating Weight-Arithmetic Safety-Drift Repair}},
  author        = {Pratinav Seth and Vinay Kumar Sankarapu},
  note          = {Under Review},
  year          = {2026}
}

All →

AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
R E Zera Marveen Lyngkhoi, Chirag Chawla, Pratinav Seth, Utsav Avaiya, Soham Bhattacharjee, Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu
Pre-Print2026Senior authorarXiv PDF
C-ΔΘ: Circuit-Restricted Weight Arithmetic for Selective Refusal
Aditya Kasliwal, Pratinav Seth, Vinay Kumar Sankarapu
Mech Interp Workshop, ICML 20262026Senior authorWorkshoparXiv PDF1 citation
ALIGNBEAM: Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing
Chirag Chawla, Pratinav Seth, Vinay Kumar Sankarapu
EMNLP 2026 Main Conference · AI for Good Workshop, ICML 20262026Senior authorConferencearXiv PDF
Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution
Saisab Sadhu, Pratinav Seth, Vinay Kumar Sankarapu
Pre-Print2026Senior authorarXiv PDF

Self-Calibrating Weight-Arithmetic Safety-Drift Repair

Cite

More in Safety Post-Training & Alignment