Research Papers

This research introduces the Decentralized Finance (DeFi) TrustBoost Framework, which combines blockchain technology and Explainable AI to address challenges faced by lenders underwriting small business loan applications from low-wealth households. The framework is designed with a strong emphasis on fulfilling four crucial requirements of blockchain and AI systems: confidentiality, compliance with data protection laws, resistance to adversarial attacks, and compliance with regulatory audits. It presents a technique for tamper-proof auditing of automated AI decisions and a strategy for on-chain (inside-blockchain) and off-chain data storage to facilitate collaboration within and across financial organizations.

Backdoor attacks pose severe security threats to deep neural networks by embedding malicious triggers that force misclassification. While machine unlearning techniques can remove backdoor behaviors, current methods lack transparency and real-time interpretability. This paper introduces a novel framework that integrates Gradient-weighted Class Activation Mapping (Grad-CAM) into the unlearning process to provide real-time monitoring and explainability. We propose the Trigger Attention Ratio (TAR) metric to quantitatively measure the model's attention shift from trigger patterns to legitimate object features. Our balanced unlearning strategy combines gradient ascent on backdoor samples, Elastic Weight Consolidation (EWC) for catastrophic forgetting prevention, and a recovery phase for clean accuracy restoration. Experiments on CIFAR-10 with BadNets attacks demonstrate that our approach reduces Attack Success Rate (ASR) from 96.51% to 5.52% while retaining 99.48% of clean accuracy (82.06%), achieving a 94.28% ASR reduction. The integration of explainable AI enables transparent, observable, and verifiable backdoor removal.

Large Language Models (LLMs) remain susceptible to jailbreak exploits that bypass safety filters and induce harmful or unethical behavior. This work presents a systematic taxonomy of existing jailbreak defenses across prompt-level, model-level, and training-time interventions, followed by three proposed defense strategies. First, a Prompt-Level Defense Framework detects and neutralizes adversarial inputs through sanitization, paraphrasing, and adaptive system guarding. Second, a Logit-Based Steering Defense reinforces refusal behavior through inference-time vector steering in safety-sensitive layers. Third, a Domain-Specific Agent Defense employs the MetaGPT framework to enforce structured, role-based collaboration and domain adherence. Experiments on benchmark datasets show substantial reductions in attack success rate, achieving full mitigation under the agent-based defense. Overall, this study highlights how jailbreaks pose a significant security threat to LLMs and identifies key intervention points for prevention, while noting that defense strategies often involve trade-offs between safety, performance, and scalability. Code is available at: https://github.com/Kuro0911/CS5446-Project

This paper introduces BRIES, a novel compound AI architecture designed to detect and measure the effectiveness of persuasion attacks across information environments. We present a system with specialized agents: a Twister that generates adversarial content employing targeted persuasion tactics, a Detector that identifies attack types with configurable parameters, a Defender that creates resilient content through content inoculation, and an Assessor that employs causal inference to evaluate inoculation effectiveness. Experimenting with the SemEval 2023 Task 3 taxonomy across the synthetic persuasion dataset, we demonstrate significant variations in detection performance across language agents. Our comparative analysis reveals significant performance disparities with GPT-4 achieving superior detection accuracy on complex persuasion techniques, while open-source models like Llama3 and Mistral demonstrated notable weaknesses in identifying subtle rhetorical, suggesting that different architectures encode and process persuasive language patterns in fundamentally different ways. We show that prompt engineering dramatically affects detection efficacy, with temperature settings and confidence scoring producing model-specific variations; Gemma and GPT-4 perform optimally at lower temperatures while Llama3 and Mistral show improved capabilities at higher temperatures. Our causal analysis provides novel insights into socio-emotional-cognitive signatures of persuasion attacks, revealing that different attack types target specific cognitive dimensions. This research advances generative AI safety and cognitive security by quantifying LLM-specific vulnerabilities to persuasion attacks and delivers a framework for enhancing human cognitive resilience through structured interventions before exposure to harmful content.

Evaluating and measuring AI Safety Level (ASL) threats are crucial for guiding stakeholders to implement safeguards that keep risks within acceptable limits. ASL-3+ models present a unique risk in their ability to uplift novice non-state actors, especially in the realm of biosecurity. Existing evaluation metrics, such as LAB-Bench, BioLP-bench, and WMDP, can reliably assess model uplift and domain knowledge. However, metrics that better contextualize "real-world risks" are needed to inform the safety case for LLMs, along with scalable, open-ended metrics to keep pace with their rapid advancements. To address both gaps, we introduce MOCET, an interpretable and doubly-scalable metric (automatable and open-ended) that can quantify real-world risks.

Ransomware continues to evolve in complexity, making early and explainable detection a critical requirement for modern cybersecurity systems. This study presents a comparative analysis of three Transformer-based Large Language Models (LLMs) (BERT, RoBERTa, and DeBERTa) for ransomware detection using two structured datasets: UGRansome and Process Memory (PM). Since LLMs are primarily designed for natural language processing (NLP), numerical and categorical ransomware features were transformed into textual sequences using KBinsDiscretizer and token-based encoding. This enabled the models to learn behavioural patterns from system activity and network traffic through contextual embeddings. The models were fine-tuned on approximately 2,500 labelled samples and evaluated using accuracy, F1 score, and ROC-AUC. To ensure transparent decision-making in this high-stakes domain, two explainable AI techniques (LIME and SHAP) were applied to interpret feature contributions. The results show that the models learn distinct ransomware-related cues: BERT relies heavily on dominant file-operation features, RoBERTa demonstrates balanced reliance on network and financial signals, while DeBERTa exhibits strong sensitivity to financial and network-traffic indicators. Visualisation of embeddings further reveals structural differences in token representation, with RoBERTa producing more isotropic embeddings and DeBERTa capturing highly directional, disentangled patterns. In general, RoBERTa achieved the strongest F1-score, while BERT yielded the highest ROC-AUC performance. The integration of LLMs with XAI provides a transparent framework capable of identifying feature-level evidence behind ransomware predictions.

Transparency and security are both central to Responsible AI, but they may conflict in adversarial settings. We investigate the strategic effect of transparency for agents through the lens of transferable adversarial example attacks. In transferable adversarial example attacks, attackers maliciously perturb their inputs using surrogate models to fool a defender's target model. These models can be defended or undefended, with both players having to decide which to use. Using a large-scale empirical evaluation of nine attacks across 181 models, we find that attackers are more successful when they match the defender's decision; hence, obscurity could be beneficial to the defender. With game theory, we analyze this trade-off between transparency and security by modeling this problem as both a Nash game and a Stackelberg game, and comparing the expected outcomes. Our analysis confirms that only knowing whether a defender's model is defended or not can sometimes be enough to damage its security. This result serves as an indicator of the general trade-off between transparency and security, suggesting that transparency in AI systems can be at odds with security. Beyond adversarial machine learning, our work illustrates how game-theoretic reasoning can uncover conflicts between transparency and security.

Post-hoc explainability methods are a subset of Machine Learning (ML) that aim to provide a reason for why a model behaves in a certain way. In this paper, we show a new black-box model-agnostic adversarial attack for post-hoc explainable Artificial Intelligence (XAI), particularly in the image domain. The goal of the attack is to modify the original explanations while being undetected by the human eye and maintain the same predicted class. In contrast to previous methods, we do not require any access to the model or its weights, but only to the model's computed predictions and explanations. Additionally, the attack is accomplished in a single step while significantly changing the provided explanations, as demonstrated by empirical evaluation. The low requirements of our method expose a critical vulnerability in current explainability methods, raising concerns about their reliability in safety-critical applications. We systematically generate attacks based on the explanations generated by post-hoc explainability methods (saliency maps, integrated gradients, and DeepLIFT SHAP) for pretrained ResNet-18 and ViT-B16 on ImageNet. The results show that our attacks could lead to dramatically different explanations without changing the predictive probabilities. We validate the effectiveness of our attack, compute the induced change based on the explanation with mean absolute difference, and verify the closeness of the original image and the corrupted one with the Structural Similarity Index Measure (SSIM).

AI systems have found a wide range of real-world applications in recent years. The adoption of edge artificial intelligence, embedding AI directly into edge devices, is rapidly growing. Despite the implementation of guardrails and safety mechanisms, security vulnerabilities and challenges have become increasingly prevalent in this domain, posing a significant barrier to the practical deployment and safety of AI systems. This paper proposes an agentic AI safety architecture that leverages 3D to integrate a dedicated safety layer. It introduces an adaptive AI safety infrastructure capable of dynamically learning and mitigating attacks against the AI system. The system leverages the inherent advantages of co-location with the edge computing hardware to continuously monitor, detect and proactively mitigate threats to the AI system. The integration of local processing and learning capabilities enhances resilience against emerging network-based attacks while simultaneously improving system reliability, modularity, and performance, all with minimal cost and 3D integration overhead.

We introduce LiveSecBench, a continuously updated safety benchmark specifically for Chinese-language LLM application scenarios. LiveSecBench constructs a high-quality and unique dataset through a pipeline that combines automated generation with human verification. By periodically releasing new versions to expand the dataset and update evaluation metrics, LiveSecBench provides a robust and up-to-date standard for AI safety. In this report, we introduce our second release v251215, which evaluates across five dimensions (Public Safety, Fairness & Bias, Privacy, Truthfulness, and Mental Health Safety.) We evaluate 57 representative LLMs using an ELO rating system, offering a leaderboard of the current state of Chinese LLM safety. The result is available at https://livesecbench.intokentech.cn/.

VoiceMorph: How AI Voice Morphing Reveals the Boundaries of Auditory Self-Recognition

Kye Shimizu, Minghan Gao, Ananya Ganesh +1

This study investigated auditory self-recognition boundaries using AI voice morphing technology, examining when individuals cease recognizing their own voice. Through controlled morphing between participants' voices and demographically matched targets at 1% increments using a mixed-methods design, we measured self-identification ratings and response times among 21 participants aged 18-64. Results revealed a critical recognition threshold at 35.2% morphing (95% CI [31.4, 38.1]). Older participants tolerated significantly higher morphing levels before losing self-recognition ($β$ = 0.617, p = 0.048), suggesting age-related vulnerabilities. Greater acoustic embedding distances predicted slower decision-making ($r \approx 0.5-0.53, p < 0.05$), with the longest response times for cloned versions of participants' own voices. Qualitative analysis revealed prosodic-based recognition strategies, universal voice manipulation discomfort, and awareness of applications spanning assistive technology to security risks. These findings establish foundational evidence for individual differences in voice morphing detection, with implications for AI ethics and vulnerable population protection as voice synthesis becomes accessible.

Large Language Models are susceptible to jailbreak attacks that bypass built-in safety guardrails (e.g., by tricking the model with adversarial prompts). We propose Concept Alignment and Concept Manipulation CALM, an inference-time method that suppresses harmful concepts by modifying latent representations of the last layer of the model, without retraining. Leveraging concept whitening technique from Computer Vision combined with orthogonal projection, CALM removes unwanted latent directions associated with harmful content while preserving model performance. Experiments show that CALM reduces harmful outputs and outperforms baseline methods in most metrics, offering a lightweight approach to AI safety with no additional training data or model fine-tuning, while incurring only a small computational overhead at inference.