Architecture Components
Security research involving model fine-tuning processes
23 matching entries out of 126 in this category
Controlled safety-tuning experiments link boilerplate refusal statements to unnecessary refusals of benign requests. Request-specific rationales improve benign compliance, with benchmark-dependent safety tradeoffs.
Source: arXiv
KoNA measures whether vision-language models answer valid image questions while refusing unsafe components or correcting unsupported premises. Its 9,300 question-answer pairs include mixed and fully answerable controls.
Source: arXiv
Third-party fine-tuning adapters may contain backdoors. Z-PEFT screens adapter weights using spectral features, evaluated on PADBench's 13,300 adapters.
Source: arXiv
Generative reward models deployed as LLM-as-a-Judge (LaaJ) evaluators contain a logic bypass vulnerability where superficial "master key" inputs trigger false positive rewards regardless of actual response quality. Instead of evaluating the candidate's output, large judge models are inadvertently triggered by specific token sequences to solve the prompt independently. This allows malicious actors or policy models undergoing reinforcement learning to consistently game the reward signal by…
Source: arXiv
Large Language Models (LLMs) exhibit a vulnerability termed "chunky post-training," where the model learns spurious correlations between incidental prompt features (e.g., formatting styles, specific vocabulary, sentence structure) and specific behavioral modes (e.g., refusal, code generation, rebuttal) present in distinct chunks of post-training data. This results in behavioral mis-routing during inference, where benign inputs sharing surface-level features with restricted or specialized…
Source: arXiv
Discrete image tokenizers are vulnerable to unsupervised embedding-space adversarial attacks. Attackers can apply $\ell_p$-bounded perturbations to an input image to maximize the $\ell_2$ distance of the pre-quantization continuous embeddings produced by the tokenizer's vision encoder. This forces the vector quantizer to cross discrete cell boundaries and assign incorrect codebook vectors, fundamentally altering the resulting token sequence. Because the attack targets the pre-quantization…
Source: arXiv
Large Language Models (LLMs) aligned via standard preference-based optimization methods (e.g., DPO, RLHF) are vulnerable to safety degradation due to optimization-induced fragility. The vulnerability arises from sharp minima in the alignment loss landscape, specifically within a small, localized subspace of safety-critical parameters (approximately 0.5% of neurons account for >80% of worst-case alignment loss). Standard alignment algorithms enforce uniform constraints or fail to control the…
Source: arXiv
Autoregressive Large Language Models (LLMs) utilizing standard fine-tuning (SFT) or alignment techniques (RLHF/DPO) are vulnerable to training-time data poisoning attacks that exploit the sequential nature of token generation. Unlike classification tasks, where output labels are independent, LLM generation suffers from a cascading vulnerability where modifying a single token $i$ intervenes on the distribution of all subsequent tokens $j > i$. An adversary can inject a small fraction of…
Source: arXiv
LLM-as-a-Judge systems utilizing natural language rubrics are vulnerable to Rubric-Induced Preference Drift (RIPD). This vulnerability allows an attacker (or a flawed optimization process) to refine evaluation rubrics such that they maintain high agreement with human references on standard validation benchmarks while inducing systematic, directional preference degradation on unseen target domains. The attack exploits the disconnect between benchmark validation and target generalization by…
Source: arXiv
Closed-loop, self-evolving Large Language Model (LLM) multi-agent systems (MAS) are vulnerable to irreversible safety erosion and alignment failure. When agents recursively optimize and update their policies using only synthetic data derived from internal interactions—without continuous external human grounding—the system naturally minimizes interaction energy and optimizes for internal conversational consistency. This isolation causes a progressive drift away from initial anthropic safety…
Source: arXiv