Hard Negatives Reveal What Easy Negatives Hide: Cross-Lingual Harmfulness Representations Degrade with Resource Tier Under Hard Negatives

arXiv:2609.27758 · cs.CL, cs.AI · Submitted 2026-08-15 · Read on arXiv

cs.CL, cs.AI

Submitted: 2026-08-15

Updated: 2026-08-15

Code: https://github.com/tatsu-lab/stanford_alpaca

Terminology

• AUROC — 34× in this paper · explained in 2 episodes
• Low-Resource Languages — 23× in this paper · explained in 4 episodes
• Calibration — 4× in this paper · explained in 7 episodes
• Alignment — 2× in this paper · explained in 1 episode
• Hallucination — 2× in this paper · explained in 3 episodes
• Regularization — 2× in this paper · explained in 3 episodes
• Steering — 2× in this paper · explained in 2 episodes
• Accuracy — 1× in this paper · explained in 1 episode
• Adversarial Attacks — 1× in this paper · explained in 1 episode
• Calibration Failure — 1× in this paper · explained in 2 episodes
• Cross-lingual Transfer — 1× in this paper · explained in 2 episodes
• Entanglement — 1× in this paper · explained in 4 episodes
• Quantization — 1× in this paper · explained in 12 episodes
• Transformer — 1× in this paper · explained in 2 episodes

Sources

• Low-Resource Safety Failures Are Action Failures, Not Representation Failures
• Do LLM hallucination detectors suffer from low-resource effect?
• Before the Last Token: Diagnosing Final-Token Safety Probe Failures
• Harmful Intent as a Geometrically Recoverable Feature of LLM Residual Streams
• The Illusion of Cross-Lingual Safety in Low-Resource Languages
• The Entanglement Wall: Activation-Space Probes as Risk Detectors, Not Context Adjudicators
• Refusal Direction is Universal Across Safety-Aligned Languages
• Low-Resource Languages Jailbreak GPT-4
• LLMs Encode Harmfulness and Refusal Separately

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering