No article text was recorded for this version.
superficial safety alignment
Version 2 of 286 papers117 conceptsfrozen 2026-10-04
Papers in this version
- Blog: The First International AI Safety Report
- Blog: Steering MoE LLMs via Expert (De)Activation
- Blog: Jet Expansions of Residual Computation
- Blog: Intrinsic Model Weaknesses: How Priming Attacks Unveil Vulnerabilities in Large Language Models
- Blog: Who's asking? User personas and the mechanics of latent misalignment
- Blog: Jailbreak Open-Sourced Large Language Models via Enforced Decoding
- Blog: JULI: Jailbreak Large Language Models by Self-Introspection
- Blog: Emulated Disalignment: Safety Alignment for Large Language Models May Backfire!
- Blog: The Safety Gap Toolkit: Evaluating Hidden Dangers of Open-Source Models
- Blog: Safety Alignment via Constrained Knowledge Unlearning
- Blog: Who's asking? User personas and the mechanics of latent misalignment
- Blog: Fundamental Limitations of Alignment in Large Language Models
- Blog: Representation Noising: A Defence Mechanism Against Harmful Finetuning
- Blog: Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications
- Blog: On the Safety of Open-Sourced Large Language Models: Does Alignment Really Prevent Them From Being Misused?
- Blog: Rapid Optimization for Jailbreaking LLMs via Subconscious Exploitation and Echopraxia
- Blog: Backdoor Activation Attack: Attack Large Language Models using Activation Steering for Safety-Alignment
- Blog: JULI: Jailbreak Large Language Models by Self-Introspection
- Blog: On the Vulnerability of Safety Alignment in Open-Access LLMs
- Blog: TRACEALIGN - Tracing the Drift: Attributing Alignment Failures to Training-Time Belief Sources in LLMs
- Blog: Linearly Decoding Refused Knowledge in Aligned Language Models
- Blog: Small Changes, Big Impact: How Manipulating a Few Neurons Can Drastically Alter LLM Aggression
- Blog: Probing the Safety Response Boundary of Large Language Models via Unsafe Decoding Path Generation
- Blog: Why Safeguarded Ships Run Aground? Aligned Large Language Models' Safety Mechanisms Tend to Be Anchored in The Template Region
- Blog: A Granular Study of Safety Pretraining under Model Abliteration
- Blog: What Makes and Breaks Safety Fine-tuning? A Mechanistic Study
- Blog: Curiosity-driven Red-teaming for Large Language Models
- Blog: No More, No Less: Least-Privilege Language Models
- Blog: MDIT-Bench: Evaluating the Dual-Implicit Toxicity in Large Multimodal Models
- Blog: The Alignment Problem in Context
- Blog: Detoxifying Large Language Models via Knowledge Editing
- Blog: Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
- Blog: Alignment Whack-a-Mole : Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Models
- Blog: Structure and Destructure: Dual Forces in the Making of Knowledge Engines
- Blog: Revealing the Intrinsic Ethical Vulnerability of Aligned Large Language Models
- Blog: Ring-A-Bell! How Reliable are Concept Removal Methods For Diffusion Models?
- Blog: Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection
- Blog: Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing
- Blog: Large Language Model Safety: A Holistic Survey
- Blog: Soft Prompt Threats: Attacking Safety Alignment and Unlearning in Open-Source LLMs through the Embedding Space
- Blog: Emulated Disalignment: Safety Alignment for Large Language Models May Backfire!
- Blog: Detoxifying Large Language Models via Knowledge Editing
- Blog: AI Safeguards as Affordance Modulation: Embedded Population Assumptions in Agentic Systems
- Blog: Improving LLM Safety Alignment with Dual-Objective Optimization
- Blog: Modular Pretraining Enables Access Control
- Blog: Bayesian scaling laws for in-context learning
- Blog: Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs
- Blog: Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs
- Blog: Towards Safer Large Language Models through Machine Unlearning
- Blog: Uncovering Safety Risks of Large Language Models through Concept Activation Vector
- Blog: Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
- Blog: Probing the Robustness of Large Language Models Safety to Latent Perturbations
- Blog: Biais de genre encodés et exprimés dans les LLM : étude conjointe
- Blog: Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks
- Blog: Dissecting Language Models: Machine Unlearning via Selective Pruning
- Blog: Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks
- Blog: Are aligned neural networks adversarially aligned?
- Blog: Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge
- Blog: Safety Pretraining: Toward the Next Generation of Safe AI
- Blog: REPO: Detoxifying LLMs via Representation Erasure-based Preference Optimization
Showing 60 of 86 papers.
History & citation
Current · Permanent link
Cite this direction
Latest version (a living page):
@misc{lacuna_superficial_safety_alignment_b57a074f,
title = {superficial safety alignment},
author = {{Lacuna}},
howpublished = {Lacuna research direction},
year = {2026},
url = {https://lacuna.tiptreesystems.com/direction/superficial-safety-alignment/rdr_b57a074f99b547f9a9d5037f80abc713},
note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}This exact version (version 2, frozen text):
@misc{lacuna_superficial_safety_alignment_b57a074f_v2,
title = {superficial safety alignment},
author = {{Lacuna}},
howpublished = {Lacuna research direction},
year = {2026},
url = {https://lacuna.tiptreesystems.com/direction/superficial-safety-alignment/rdr_b57a074f99b547f9a9d5037f80abc713/v/2},
note = {Version 2, frozen 2026-10-04. Accessed 2026-11-21},
}