arXiv 2601.12639 v1, January 19, 2026. University of Cincinnati / University of Toronto / University of Oxford. Controlled six-objective fine-tuning comparison — SFT, DPO, Conditional Fine-Tuning (Korbak et al. 2023), Inoculation Prompting (Tan et al. 2025, Wichers et al. 2025), Odds Ratio Preference Optimization (Hong et al. 2024), and KL-regularized fine-tuning — holding training data, domain, LoRA architecture, and optimization fixed, on LLaMA-3.1-8B-Instruct (primary) with Gemma2-2B, Gemma2-9B, Qwen3-4B, and Qwen2.5-7B replication. Four domains (GSM8K, SuperGPQA-engineering, legal reasoning, cybersecurity); three evaluation axes (capability via task accuracy or LLM-judge rubric, adversarial robustness via five StrongREJECT prompting jailbreaks, persona drift via Dark Triad probes from Perez et al. 2022 "Discovering language model behaviors with model-written evaluations"). Headline finding: at small training budgets (25k–50k tokens) safety outcomes are similar across objectives and capability is the dominant axis of variation; at large budgets (200k–800k tokens) objectives diverge — SFT and DPO couple capability gains to monotonic increases in adversarial vulnerability and Dark Triad endorsement; ORPO and KL-regularized fine-tuning show no statistically significant persona drift at any scale; Inoculation Prompting is Pareto-efficient on robustness (matches SFT capability while substantially suppressing ASR growth) but matches SFT on persona drift.