Automatically jailbreaks LLMs with black-box access.
problem Generating harmful content from black-box LLMs.
method Automated method using an attacker LLM to refine prompts.
result Generates jailbreaks for over 80% of prompts.
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,742 papers · 148 categories
Trend · papers per month
19 results for “Jailbreaking”
VERA uses variational inference to jailbreak LLMs without manual optimization.
problem Lack of principled objective for gradient-based optimization in jailbreaking LLMs.
method VERA casts black-box jailbreak prompting as a variational inference problem, training a small attacker LLM to approximate the target LLM's posterior over adversarial prompts.
result VERA achieves strong performance across various target LLMs, demonstrating the value of probabilistic inference for adversarial prompt generation.
SmoothLLM defends LLMs from jailbreaking attacks by randomly perturbing inputs.
problem Adversaries can fool large language models into generating objectionable content.
method SmoothLLM randomly perturbs multiple copies of a prompt and aggregates predictions to detect adversarial inputs.
result SmoothLLM sets the state-of-the-art for robustness against various jailbreak attacks.
Shorter adversarial prompts help protect LLMs from jailbreak attacks.
problem Protecting large language models from jailbreak attacks with long adversarial suffixes.
method Adversarial training on shorter adversarial suffixes to defend against longer adversarial suffixes.
result Aligning LLMs on shorter adversarial suffixes can effectively defend against jailbreak attacks with longer suffixes.
Simple adaptive attacks can jailbreak state-of-the-art safety-aligned language models.
problem Vulnerability of safety-aligned language models to simple adaptive attacks.
method Designing adversarial prompt templates, random search on suffixes, and transfer/prefilling attacks.
result 100% attack success rate on various models including GPT-4, Vicuna, Mistral, and Claude.
RESTA defends LLMs against jailbreaking attacks by adding random noise to embeddings.
problem Vulnerability of LLMs to jailbreaking attacks that generate harmful outputs.
method Adds random noise to embedding vectors and aggregates during token generation.
result RESTA achieves superior robustness versus utility tradeoffs compared to baseline defenses.
Understanding and Improving Continuous Adversarial Training for LLMs via In-context Learning Theorycs.LG
This paper improves continuous adversarial training for LLMs using in-context learning theory.
problem Efficiently defending large language models (LLMs) against jailbreak attacks.
method The paper presents a theoretical analysis of continuous adversarial training (CAT) for LLMs based on in-context learning (ICL) theory, proving a robust generalization bound and proposing an improved regularization term.
result The robust generalization bound explains why CAT can defend against jailbreak prompts and shows that LLM robustness is related to embedding matrix singular values.
VERA-V uses variational inference to discover vulnerabilities in multimodal vision-language models.
problem Existing methods for jailbreaking vision-language models are brittle, limited, and focus on single attacks.
method VERA-V recasts jailbreak discovery as learning a joint posterior distribution over text-image prompts, using variational inference and three complementary strategies.
result VERA-V consistently outperforms state-of-the-art baselines, achieving up to 53.75% higher attack success rate.
COCA refactors training data to identify and erase unsafe concepts in LLMs.
problem Identifying and erasing unsafe concepts in Large Language Models (LLMs) for safety alignment.
method Concept Concentration (COCA) refactors training data with an explicit reasoning process to identify and erase unsafe concepts.
result COCA significantly reduces both in-distribution and out-of-distribution jailbreak success rates while maintaining strong performance on regular tasks.
High-dimensional models pose both safety benefits and risks.
problem Emergent problems in safety alignment due to high-dimensional representations.
method Detailed visualizations and lower-dimensional subspace projections.
result Dimensional reduction preserves safety alignment while avoiding linear jailbreaking.
The paper proves neural networks are almost always surjective, impacting model safety.
problem Ensuring neural networks can generate any output, including harmful content.
method Analyzing fundamental neural architectures and generative models.
result Many neural architectures are almost always surjective, allowing for arbitrary outputs.
Diffusion LLMs can efficiently generate harmful prompts for adversarial testing.
problem Generating harmful prompts for adversarial testing is resource-intensive and costly.
method Transformed adversarial prompt optimization into an efficient inference task using pretrained Diffusion LLMs.
result Only a few conditional samples are required to generate harmful prompts with high reward.
New framework assesses LLM security risks in BFSI.
problem Lack of domain-specific security evaluation for LLMs in BFSI.
method Risk-aware evaluation framework combining taxonomy, automated red-teaming, and ensemble judging.
result Higher decoding stochasticity and adaptive interaction lead to more severe disclosures.
Theoretical analysis shows LLMs can self-correct responses through in-context learning.
problem Understanding how large language models improve through self-correction.
method Theoretical analysis based on simplified alignment task, focusing on softmax attention, multi-head attention, and MLP blocks.
result LLMs can refine responses in an in-context way when given accurate self-examinations as rewards.
This paper formalizes AI safety using hypothesis testing in GenAI.
problem Ensuring safety of generative AI tools that create realistic content.
method Formalization of computational safety through hypothesis testing and signal processing.
result Demonstrates how AI safety can be assessed quantitatively using mathematical frameworks.
ELS framework improves safety alignment by dynamically steering LLMs towards helpful responses.
problem Over-Refusal in Aligned Large Language Models
method Fine-tuning free framework using an Energy-Based Model (EBM) to dynamically steer LLMs during inference.
result Extensive experiments show a significant reduction in false refusals (from 57.3% to 82.6%) while maintaining safety performance.
Online monitoring system for safety classifiers with shift detection and conformal adaptation
problem Detecting and adapting to distributional shifts in deployed safety classifiers
method Calibrated sequential statistics for online monitoring, conformal abstention for adaptation
result 86.6% valid detection with mean latency of 39.5 steps
New method accounts for hidden context in preference learning for RLHF models.
problem Incomplete data with hidden context affects RLHF model outcomes.
method Distributional Preference Learning (DPL) methods estimate hidden context distributions.
result DPL methods reduce RLHF vulnerabilities by accounting for hidden context.
Online monitor detects classifier drift and adapts predictions.
problem Silent degradation of classifier accuracy under distributional shift.
method Sliding-window KS statistic with calibrated alarm thresholds.
result 86.6% valid detection across various shift conditions.