Research
Academic papers, new techniques, benchmarks, and theoretical findings in AI/LLM security.
Academic papers, new techniques, benchmarks, and theoretical findings in AI/LLM security.
2,033 items
This design study reports on building a data space for the cultural sector, where actors differ widely in size, structure, digital maturity and technical capability. The authors designed and qualitatively evaluated two foundational artifacts, a network architecture and a data exchange standard, to balance interoperability with those differing capabilities. They derive a design principle: data spaces should include heterogeneous actors by simplifying participation and lowering entry barriers, with fair access regardless of size, resources or technical capability.
This research paper argues that Information Systems scholars have largely overlooked technology-facilitated domestic and family abuse (TFDFA). It cites ANROWS findings that one in two Australian adults report victimisation through technology-facilitated abuse and one in four admit perpetration. The authors call for an inclusive IS research agenda that develops frameworks, interventions and design principles to mitigate abuse.
A Professional Development Workshop at the Academy of Management 2025 meeting in Copenhagen examined twin transition, the alignment of digital and sustainability transformations. The workshop reported five recurring themes: conceptual ambiguity, metrics and data, rhetoric-practice gaps, politics and power, and researcher reflexivity. The authors identify four pitfalls that threaten progress: technological determinism, ideological overconfidence, fragmentation and analysis-paralysis.
Researchers Muhammad Khalid Anser, Sajid Ali, Yi Chen and Raima Nazar examine cybercrime during pandemic-era uncertainty. The source text provided here contains only the publication metadata (Computers & Security, available online 15 May 2026), so the study's method and findings cannot be stated from it.
Yan, Yang, Cao, Meng and Sun, in the Journal of Information Security and Applications (July 2026), study auditable data sharing across domains. The source text provided gives no method, findings or numbers beyond the title and bibliographic details.
This paper proposes a policy-based conjunctive scheme for digital forgetting of co-owned data. It appears in ACM Transactions on Privacy and Security, Volume 29, Issue 3, pages 1-37, August 2026. The source text provides no further details on its method or findings.
Zhe Jiang, Kai Zhang, Junqing Gong and Haifeng Qian published a paper on laconic attribute-based private set intersection (PSI) on authenticated inputs and applications in the Journal of Information Security and Applications, Volume 100, in July 2026. The source text provided contains only publication metadata, so the research question, method and findings cannot be described from it.
Yuan, Chen, Zhao and Zhang published an empirical study in Computers & Security, available online 14 May 2026, examining attack mitigations in compiled Rust and Go binaries. The source text provided here contains only the publication metadata and author list, so the research question, method and findings cannot be stated from it.
The authors present FastBOC, a hybrid covert communication framework that pairs a blockchain, used as a covert signal channel for lightweight signals, with an onion network that carries high-capacity secret data. They implemented it on the Ethereum testnet and report that it transmits 1-Megabyte (MB) data within 20.11 seconds and reduces communication cost by 5–7 orders of magnitude compared to prior blockchain-based covert communication schemes.
The paper proposes a twin driver approach in which paired TEE and REE drivers cooperate to give trusted execution environments such as TrustZone secure, efficient driver support. It introduces DrvAna, an LLVM-based data flow analysis framework that uses a value-type tree to analyze shared driver state, and a Linux driver runtime (LDR) that sandboxes the TEE driver using ARM domain access control. Adapting six existing Linux drivers to LDR on an NXP IMX6Q SABRE-SD board, the authors report performance comparable to Linux with negligible overhead.
A July 2026 paper in the Journal of Information Security and Applications, Volume 100, by Xiaojuan Dong, Shu Wu, Yadong Yang and Bao Zhou, presents MPV, a scheme for restricting master key decryption authority in multi-user Paillier systems using mixed ciphertexts. The source text provided contains only the publication metadata and no method details or findings.
An ACPHIS-convened panel of Australian information systems academics examined institutional readiness, policy gaps and opportunities for generative AI in higher education at ACIS 2025. The panel drew on interim findings from an ACPHIS-sponsored national project involving 45 public universities across Australia and New Zealand, along with systematic literature reviews and policy analyses. Key themes include responsible AI use, graduate skill development, learning assurance, privacy, intellectual property and security.
A panel report from a World Café symposium at the European Academy of Management 2025 Conference examines how academics experience AI reshaping teaching, research and academic service. Participants saw opportunities to boost teaching and research productivity but also voiced concern about ethical and professional risks. The authors propose a research agenda for Information System scholars covering human–AI collaboration, responsible and trustworthy AI tools, cross-context adoption patterns and career effects for early-career academics.
The source is a survey titled "Towards Robust and Secure Embodied AI: A Survey on Vulnerabilities and Attacks," published in ACM Computing Surveys, Volume 58, Issue 12, pages 1-36, September 2026. The provided text contains only the publication details and no article content describing its findings or methods.
PVLM is a parsing-aware vision-language model with dynamic contrastive learning for zero-shot deepfake attribution, meaning it traces forged faces to generators not seen in training, including diffusion models. The authors build a fine-grained ZS-DFA benchmark and use face parsing to exploit differences in how GAN and diffusion generators preserve source facial attributes. They report that the model exceeds the state of the art on the ZS-DFA benchmark across various protocol evaluations.
The authors introduce a taxonomy and benchmarking framework that classifies deepfake attacks by intent and generative mechanism, then evaluate them against deep face recognition systems. Diverse deepfake attacks commonly exceed 70% success and, in some regimes, surpass 90%. The study finds that attack success depends on the degree of identity controllability rather than visual quality.
Cisco researchers disclosed MemoryTrap, a Claude Code vulnerability in which a routine workflow (cloning a repository, letting the agent help, approving an npm dependency installation) let a malicious payload reach persistent memory, the global hooks configuration, and the system prompt. The corrupted context then shaped the model's behavior across sessions, projects and reboots, which the article maps to OWASP ASI06: Memory & Context Poisoning.
Fix: Claude Code v2.1.50 removed user memories from the system prompt, reducing the specific high-trust override path identified. The source notes this fixes only that path and calls for treating memory as part of the attack surface, without specifying further controls.
OWASP GenAI Security