Research Papers

Jailbreak prompts pose a significant threat in AI and cybersecurity, as they are crafted to bypass ethical safeguards in large language models, potentially enabling misuse by cybercriminals. This paper analyzes jailbreak prompts from a cyber defense perspective, exploring techniques like prompt injection and context manipulation that allow harmful content generation, content filter evasion, and sensitive information extraction. We assess the impact of successful jailbreaks, from misinformation and automated social engineering to hazardous content creation, including bioweapons and explosives. To address these threats, we propose strategies involving advanced prompt analysis, dynamic safety protocols, and continuous model fine-tuning to strengthen AI resilience. Additionally, we highlight the need for collaboration among AI researchers, cybersecurity experts, and policymakers to set standards for protecting AI systems. Through case studies, we illustrate these cyber defense approaches, promoting responsible AI practices to maintain system integrity and public trust. \textbf{\color{red}Warning: This paper contains content which the reader may find offensive.}

Explainable AI~(XAI) methods such as SHAP can help discover feature attributions in black-box models. If the method reveals a significant attribution from a ``protected feature'' (e.g., gender, race) on the model output, the model is considered unfair. However, adversarial attacks can subvert the detection of XAI methods. Previous approaches to constructing such an adversarial model require access to underlying data distribution, which may not be possible in many practical scenarios. We relax this constraint and propose a novel family of attacks, called shuffling attacks, that are data-agnostic. The proposed attack strategies can adapt any trained machine learning model to fool Shapley value-based explanations. We prove that Shapley values cannot detect shuffling attacks. However, algorithms that estimate Shapley values, such as linear SHAP and SHAP, can detect these attacks with varying degrees of effectiveness. We demonstrate the efficacy of the attack strategies by comparing the performance of linear SHAP and SHAP using real-world datasets.

The rapid proliferation of open-source language models significantly increases the risks of downstream backdoor attacks. These backdoors can introduce dangerous behaviours during model deployment and can evade detection by conventional cybersecurity monitoring systems. In this paper, we introduce a novel class of backdoors in transformer models, that, in contrast to prior art, are unelicitable in nature. Unelicitability prevents the defender from triggering the backdoor, making it impossible to properly evaluate ahead of deployment even if given full white-box access and using automated techniques, such as red-teaming or certain formal verification methods. We show that our novel construction is not only unelicitable thanks to using cryptographic techniques, but also has favourable robustness properties. We confirm these properties in empirical investigations, and provide evidence that our backdoors can withstand state-of-the-art mitigation strategies. Additionally, we expand on previous work by showing that our universal backdoors, while not completely undetectable in white-box settings, can be harder to detect than some existing designs. By demonstrating the feasibility of seamlessly integrating backdoors into transformer models, this paper fundamentally questions the efficacy of pre-deployment detection strategies. This offers new insights into the offence-defence balance in AI safety and security.

Data forms the backbone of artificial intelligence (AI). Privacy and data protection laws thus have strong bearing on AI systems. Shielded by the rhetoric of compliance with data protection and privacy regulations, privacy-preserving techniques have enabled the extraction of more and new forms of data. We illustrate how the application of privacy-preserving techniques in the development of AI systems--from private set intersection as part of dataset curation to homomorphic encryption and federated learning as part of model computation--can further support surveillance infrastructure under the guise of regulatory permissibility. Finally, we propose technology and policy strategies to evaluate privacy-preserving techniques in light of the protections they actually confer. We conclude by highlighting the role that technologists could play in devising policies that combat surveillance AI technologies.

Distributed trust is a nebulous concept that has evolved from different perspectives in recent years. While one can attribute its current prominence to blockchain and cryptocurrency, the distributed trust concept has been cultivating progress in federated learning, trustworthy and responsible AI in an ecosystem setting, data sharing, privacy issues across organizational boundaries, and zero trust cybersecurity. This paper will survey the concept of distributed trust in multiple disciplines. It will take a system/software architecture point of view to look at trust redistribution/shift and the associated tradeoffs in systems and applications enabled by distributed trust technologies.

Privacy and Transparency in Graph Machine Learning: A Unified Perspective

Megha Khosla

Graph Machine Learning (GraphML), whereby classical machine learning is generalized to irregular graph domains, has enjoyed a recent renaissance, leading to a dizzying array of models and their applications in several domains. With its growing applicability to sensitive domains and regulations by governmental agencies for trustworthy AI systems, researchers have started looking into the issues of transparency and privacy of graph learning. However, these topics have been mainly investigated independently. In this position paper, we provide a unified perspective on the interplay of privacy and transparency in GraphML. In particular, we describe the challenges and possible research directions for a formal investigation of privacy-transparency tradeoffs in GraphML.

Adversarial patches are images designed to fool otherwise well-performing neural network-based computer vision models. Although these attacks were initially conceived of and studied digitally, in that the raw pixel values of the image were perturbed, recent work has demonstrated that these attacks can successfully transfer to the physical world. This can be accomplished by printing out the patch and adding it into scenes of newly captured images or video footage. In this work we further test the efficacy of adversarial patch attacks in the physical world under more challenging conditions. We consider object detection models trained on overhead imagery acquired through aerial or satellite cameras, and we test physical adversarial patches inserted into scenes of a desert environment. Our main finding is that it is far more difficult to successfully implement the adversarial patch attacks under these conditions than in the previously considered conditions. This has important implications for AI safety as the real-world threat posed by adversarial examples may be overstated.

Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses

Xiao Li, Xiang Zheng, Yifeng Gao +35

Embodied Artificial Intelligence (Embodied AI) integrates perception, cognition, planning, and interaction into agents that operate in open-world, safety-critical environments. As these systems gain autonomy and enter domains such as transportation, healthcare, and industrial or assistive robotics, ensuring their safety becomes both technically challenging and socially indispensable. Unlike digital AI systems, embodied agents must act under uncertain sensing, incomplete knowledge, and dynamic human-robot interactions, where failures can directly lead to physical harm. This survey provides a comprehensive and structured review of safety research in embodied AI, examining attacks and defenses across the full embodied pipeline, from perception and cognition to planning, action and interaction, and agentic system. We introduce a multi-level taxonomy that unifies fragmented lines of work and connects embodied-specific safety findings with broader advances in vision, language, and multimodal foundation models. Our review synthesizes insights from over 500 papers spanning adversarial, backdoor, jailbreak, and hardware-level attacks; attack detection, safe training and robust inference; and risk-aware human-agent interaction. This analysis reveals several overlooked challenges, including the fragility of multimodal perception fusion, the instability of planning under jailbreak attacks, and the trustworthiness of human-agent interaction in open-ended scenarios. By organizing the field into a coherent framework and identifying critical research gaps, this survey provides a roadmap for building embodied agents that are not only capable and autonomous but also safe, robust, and reliable in real-world deployment.

Generative social robots (GSRs) powered by large language models enable adaptive, conversational tutoring but also introduce risks such as misinformation, overreliance, and privacy violations. Existing frameworks for educational technologies and responsible AI primarily define desired behaviors, yet they rarely specify the knowledge prerequisites that enable generative agents to express these behaviors reliably. To address this gap, we adopt a knowledge-based design perspective and investigate what information tutoring-oriented GSRs require to function responsibly and effectively in higher education. Based on twelve semistructured interviews with university students and lecturers, we identified twelve design requirements across three knowledge types: self-knowledge (assertive, conscientious, and friendly personality with customizable role), user-knowledge (personalized information about student learning goals, learning progress, motivation type, emotional state, and background), and context-knowledge (learning materials, educational strategies, courserelated information, and physical learning environment). Drawing from these results, this work provides a structured foundation for the design of tutoring GSRs, aligning generative AI capabilities with pedagogical and ethical expectations.

The increasing integration of AI agents into cyber-physical systems (CPS) introduces new security risks that extend beyond traditional cyber or physical threat models. Recent advances in generative AI enable deepfake and semantic manipulation attacks that can compromise agent perception, reasoning, and interaction with the physical environment, while emerging protocols such as the Model Context Protocol (MCP) further expand the attack surface through dynamic tool use and cross-domain context sharing. This survey provides a comprehensive review of security threats targeting AI agents in CPS, with a particular focus on environmental interactions, deepfake-driven attacks, and MCP-mediated vulnerabilities. We organize the literature using the SENTINEL framework, a lifecycle-aware methodology that integrates threat characterization, feasibility analysis under CPS constraints, defense selection, and continuous validation. Through an end-to-end case study grounded in a real-world smart grid deployment, we quantitatively illustrate how timing, noise, and false-positive costs constrain deployable defenses, and why detection mechanisms alone are insufficient as decision authorities in safety-critical CPS. The survey highlights the role of provenance- and physics-grounded trust mechanisms and defense-in-depth architectures, and outlines open challenges toward trustworthy AI-enabled CPS.

This paper presents an overview of the Recognize the Unseen: Unusual Behavior Recognition from Pose Data Challenge, hosted at ISAS 2025. The challenge aims to address the critical need for automated recognition of unusual behaviors in facilities for individuals with developmental disabilities using non-invasive pose estimation data. Participating teams were tasked with distinguishing between normal and unusual activities based on skeleton keypoints extracted from video recordings of simulated scenarios. The dataset reflects real-world imbalance and temporal irregularities in behavior, and the evaluation adopted a Leave-One-Subject-Out (LOSO) strategy to ensure subject-agnostic generalization. The challenge attracted broad participation from 40 teams applying diverse approaches ranging from classical machine learning to deep learning architectures. Submissions were assessed primarily using macro-averaged F1 scores to account for class imbalance. The results highlight the difficulty of modeling rare, abrupt actions in noisy, low-dimensional data, and emphasize the importance of capturing both temporal and contextual nuances in behavior modeling. Insights from this challenge may contribute to future developments in socially responsible AI applications for healthcare and behavior monitoring.

Large Language Models (LLMs) have developed rapidly in web services, delivering unprecedented capabilities while amplifying societal risks. Existing works tend to focus on either isolated jailbreak attacks or static defenses, neglecting the dynamic interplay between evolving threats and safeguards in real-world web contexts. To mitigate these challenges, we propose ACE-Safety (Adversarial Co-Evolution for LLM Safety), a novel framework that jointly optimize attack and defense models by seamlessly integrating two key innovative procedures: (1) Group-aware Strategy-guided Monte Carlo Tree Search (GS-MCTS), which efficiently explores jailbreak strategies to uncover vulnerabilities and generate diverse adversarial samples; (2) Adversarial Curriculum Tree-aware Group Policy Optimization (AC-TGPO), which jointly trains attack and defense LLMs with challenging samples via curriculum reinforcement learning, enabling robust mutual improvement. Evaluations across multiple benchmarks demonstrate that our method outperforms existing attack and defense approaches, and provides a feasible pathway for developing LLMs that can sustainably support responsible AI ecosystems.