PH, China hold bilateral talks ahead of ASEAN COC meetings
(1st UPDATE) The Philippine Department of Foreign Affairs says the two countries 'underscored the importance of sustained diplomatic dialogue and continued cooperation across all sectors'
(1st UPDATE) The Philippine Department of Foreign Affairs says the two countries 'underscored the importance of sustained diplomatic dialogue and continued cooperation across all sectors'
Prompt injection evaluations typically treat refusal as a stable, binary indicator of safety. This study challenges that paradigm by modeling refusal as a local decision boundary and examining its stability under structured perturbations. We evaluated two models, GPT-4.1 and GPT-4o, using 3,274 perturbation runs derived from refusal-inducing prompt injection attempts. Each base prompt was subjected to 25 perturbations across five structured families, with outcomes manually…
Based on one year of anthropological fieldwork in a group of villages in the Haa Valley of western Bhutan, this ethnographic paper examines how Bhutanese migrants in Australia maintain connections to their home villages through food. Although migration to Australia has become a prominent phenomenon in Bhutan (and across the Himalayan region), the paper challenges mainstream accounts of rural depopulation that portray young Bhutanese as progressively detaching from their…
Southeast Asia is quietly entering a new strategic era, one defined not by the size of a navy’s fleet, but by the range of its missiles. Across the region, an emerging pattern is becoming visible. Vietnam, the Philippines, and Indonesia are each moving rapidly to establish anti-access/area-denial (A2/AD) postures. For some, the shift is born
Large Language Models (LLMs) have enabled the development of powerful agentic systems capable of automating complex workflows across various fields. However, these systems are highly vulnerable to indirect prompt injection attacks, where malicious instructions embedded in external data can hijack agent behavior. In this work, we present ReasAlign, a model-level solution to improve safety alignment against indirect prompt injection attacks. The core idea of ReasAlign is to…
We investigate a failure mode of large language models (LLMs) in which plain-text prompts elicit excessive outputs, a phenomenon we term Overflow. Unlike jailbreaks or prompt injection, Overflow arises under ordinary interaction settings and can lead to elevated serving cost, latency, and cross-user performance degradation, particularly when scaled across many requests. Beyond usability, the stakes are economic and environmental: unnecessary tokens increase per-request cost…
Higher education continues to encounter the challenge of redesigning writing pedagogy beyond the rapid adoption of emerging technologies. This challenge is particularly evident in English writing courses, which play a role in developing students’ writing and research skills in universities across the United Arab Emirates (UAE). While generative artificial intelligence (GenAI) tools offer practical affordances for writing instruction, their growing use has also raised…
This paper presents a comprehensive empirical study on the safety alignment capabilities. We evaluate what matters for safety alignment in LLMs and LRMs to provide essential insights for developing more secure and reliable AI systems. We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques. Our large-scale evaluation is conducted using 32 recent, popular LLMs and LRMs across thirteen…
This study examines how shocks in Colombia’s cocaine economy and national enforcement dynamics in Ecuador shape homicide rates across Ecuadorian provinces. Using a balanced provincial panel (2014–2023) and two Bartik-type exposure measures, it translates national coca cultivation in Colombia and national drug-trafficking arrests in Ecuador into subnational-level shocks based on each province’s structural exposure to narcotrafficking routes, particularly proximity to…
Abstract This paper examines variation in compliance with liberal international law across Southeast Asian territorial disputes. It argues that the constraining effects of legal rulings depend on the interaction between power asymmetry and the political costs of compliance. Drawing on four cases, Pedra Branca, Sipadan-Ligitan, Preah Vihear, and the South China Sea arbitration, the paper identifies a graded pattern: law has greater effect when power asymmetry is limited and…
The nuclear industry is distinguished by its highly secured and restricted nature, requiring rigorous safety validation, which necessitates additional considerations for Artificial Intelligence (AI) adoption compared to other industries. Nuclear export control requires comprehensive assessment of strategic items through expert classification but faces growing challenges in knowledge transfer and consistent decision making across increasing international trade volumes. This…
This study examines how Christian nationalism—an ideology advocating for a specific form of conservative Christianity to be privileged in American civic life—mediates the relationship between religio-political conservatism and opposition to legal abortion across social-sphere circumstances (financial reasons, marital status, family size) and physical-sphere circumstances (rape, health risks, fetal defects) in the United States. Using structural equation modeling with 2021…
Earlier this year, during RSAC 2025, we introduced something the industry had never seen before: an open-source tool capable of generating an AI Software Bill of Materials (AIBOM) for models on Hugging Face. That launch ignited a wave of interest across the security, AI, and software supply chain communities. It confirmed what many of us
Agentic AI introduces security vulnerabilities that traditional LLM safeguards fail to address. Although recent work by Unit 42 at Palo Alto Networks demonstrated that ChatGPT-4o successfully executes attacks as an agent that it refuses in chat mode, there is no comparative analysis in multiple models and frameworks. We conducted the first systematic penetration testing and comparative evaluation of agentic AI systems, testing five prominent models (Claude 3.5 Sonnet, Gemini…
Browser-using agents (BUAs) are an emerging class of AI agents that interact with web browsers in human-like ways, including clicking, scrolling, filling forms, and navigating across pages. While these agents help automate repetitive online tasks, they are vulnerable to prompt injection attacks that trick an agent into performing undesired actions, such as leaking private information or issuing unintended state-changing requests. We propose ceLLMate, a browser-level…
With a surge of hidden electronic warfare upgrades, China is reshaping the South China Sea into an electronic battlespace tilted decisively in its favor. This month, the Asia Maritime Transparency Initiative (AMTI) reported that China has quietly expanded its electronic warfare and surveillance infrastructure across its artificial island bases in the disputed Spratly Islands. The
Introducing the OWASP Top 10 for Agentic AI Applications - our community’s actionable framework for securing autonomous, tool-using AI systems. Built at global scale informed by real incidents, and based on our work already adopted across industry, this release marks a pivotal moment in turning insight into action, and advancing the security of Agentic Applications at the pace of innovation.
This Methods article details a survey used to create spatial maps of perceived violence and danger, key indicators of criminal governance, across three Mexican states. This method adds to traditional mapping techniques, which typically only visualize areal data at the municipal level. Criminal governance has been defined as the process by which organized crime groups impose control over territory, population, and resources, often through violence, threats, and bribery. The…
The emerging "agentic web" envisions large populations of autonomous agents coordinating, transacting, and delegating across open networks. Yet many agent communication and commerce protocols treat agents as low-cost identities, despite the empirical reality that LLM agents remain unreliable, hallucinated, manipulable, and vulnerable to prompt-injection and tool-abuse. A natural response is "agents-at-stake": binding economically meaningful, slashable collateral to…
China has carried out major upgrades to its electronic warfare and surveillance systems across its major bases in the Spratly Islands to strengthen its intelligence capabilities in the South China Sea, according to an American think tank. Satellite imagery analyzed by the Asia Maritime Transparency Initiative in a report last week showed new and enhanced ...
For nearly a century, the practice of city twinning has expanded across every inhabited continent, yet it remains largely undertheorized in geography and urban studies. Questions surrounding its global proliferation and implications become more urgent as twinning expands into new geographies, such as within Asia. This study seeks to advance understanding of city diplomacy by offering insights from emerging intra-Asian twinning. Drawing on qualitative research on two sister…
In Dictatorship across borders, Mila Burns explores the Brazilian dimension of the conspiracy to topple the socialist government of Salvador Allende in Chile (1970–73). In much of the standard literature on the subject, this overthrow has been ascribed almost solely to the United States' machinations. This is not the first attempt to posit a Brazilian connection; indeed, as early as January 1974, journalist Marlise Simons drew a link between the Chilean coup and…
This stuy examines the relationship between China’s investment in Latin America and Twitter discussions, focusing on Spanish-language conversations in 16 countries. It identifies 34 topics, which come together as five frames: foreign affairs, economy, internal crisis, health, and culture. We find that higher Chinese investment correlates with more negative Twitter discussions, with unique differences across frames, challenging the idea that financial investment alone can…
This study evaluates how satisfied users are with collection management practices in business school libraries across Vietnam and China. Using a purposive sample of 401 teaching staff—lecturers, researchers, associate and full professors—from 27 institutions, all with at least three years of library use and active engagement, the research gathered data via a semi-structured questionnaire. Results indicate that current practices fall short of meeting users’ information needs…