Artificial intelligence (AI) applications have become ubiquitous in their impact on individuals and society, highlighting a crucial need for their responsible development. Recent research has called for participatory AI auditing, empowering individuals without AI expertise to audit AI applications throughout the entire AI development pipeline. Our work focuses on investigating how to support these kinds of auditors through participatory AI auditing tools and processes. We conducted a series of co-design workshops, using two health-related predictive AI applications as examples. Our results show that participants wanted to be part of AI audits, and were insightful in identifying the potential impacts of applications, but needed to be assisted in conducting audits, especially how to measure impacts. Importantly, participants provided examples of impacts not considered in current risk/harm taxonomies. Our findings provide implications for the design of tools and processes to empower everyone to contribute to responsible AI development in the future.

In an era characterised by evolving cyber threats and sophisticated adversar-ial behaviour, the field of cyber-security faces a continuous and formidablechallenge. The development of dynamic and adaptive security control mea-sures is imperative in order to safeguard critical assets and information.This article delves into the realm of Moving Target Defence (MTD), astrategic approach that seeks to outmanoeuvre adversaries by constantlyshifting the security landscape. Our research specifically focuses on the ap-plication of Reinforcement Learning (RL) in MTD, with a focus on threatexposure and the efficacy of control strategies with respect to risk reduction.A defensive RL agent is proposed that incorporates attack graphs as ablueprint to assess possible paths that an adversary may take. By consideringreceived events about an adversary’s actions, the defensive agent continuouslyupdates its knowledge about the adversary’s position on the attack graph.The proposed research establishes and evaluates a risk-based, RL-drivenagent capable of MTD operations in order to address adversarial behaviours.The proposed approach provides valuable insights into optimally deployingsecurity controls under dynamic threat scenarios and restricted budget resources.

In this paper,..........

As we transition from Narrow Artificial Intelligence towards Artificial Super Intelligence, users are increasingly concerned about their privacy and the trustworthiness of machine learning (ML) technology. A common denominator for the metrics of trustworthiness is the quantification of uncertainty inherent in DL algorithms, and specifically in the model parameters, input data, and model predictions. One of the common approaches to address privacy-related issues in DL is to adopt distributed learning such as federated learning (FL), where private raw data is not shared among users. Despite the privacy-preserving mechanisms in FL, it still faces challenges in trustworthiness. Specifically, the malicious users, during training, can systematically create malicious model parameters to compromise the models’ predictive and generative capabilities, resulting in high uncertainty about their reliability. To demonstrate malicious behaviour, we propose a novel model poisoning attack method named Delphi which aims to maximise the uncertainty of the global model output. We achieve this by taking advantage of the relationship between the uncertainty and the model parameters of the first hidden layer of the local model. Delphi employs two types of optimisation, Bayesian Optimisation and Least Squares Trust Region, to search for the optimal poisoned model parameters, named as Delphi-BO and Delphi-LSTR. We quantify the uncertainty using the KL Divergence to minimise the distance of the predictive probability distribution towards an uncertain distribution of model output. Furthermore, we establish a mathematical proof for the attack effectiveness demonstrated in FL. Numerical results demonstrate that Delphi-BO induces a higher amount of uncertainty than Delphi-LSTR highlighting vulnerability of FL systems to model poisoning attacks.

Metaverse is envisioned to be a human-centric framework, and provide a new concept of living by offering comprehensively immersive experience for users in education, medicine and entertainment domain. Since a large amount of private data is generated at each user for accessing Metaverse, the emerging federated learning (FL) provides an effective solution to address the potential privacy leakage of data sharing by adopting the mechanism of local training and global model aggregation. However, the model aggregation is susceptible to adversarial poisoning attacks. This imposes critical issues for the privacy-preserving mechanism in Metaverse. In this research, we develop two poisoning attacks in order to emulate the behaviour of adversaries possibly existing in practical Metaverse scenarios. First, we develop a data poisoning attack using Bayesian optimisation to search for the optimal parameters of generating adversarial examples to conduct reversed adversarial training. Second, we develop a model poisoning attack where we apply layer optimisation using Bayesian optimisation to search the optimal weights for the convolutional layer in order to induce uncertainty in the classification. Numerical results show that both attack schemes can cause attacks that can not be recognised by the FL server, and layer optimisation is a stronger poisoning attack.