Nadia Vale covers security failures, system safeguards and the difficult business of establishing trust. Her beat connects technical disclosures with their consequences for users and institutions. She favors specific threat models over sweeping claims that a system is safe or unsafe.
OpenAI today executed a strategic maneuver on two fronts: introducing an “Agentic Commerce Protocol” within ChatGPT for enhanced product discovery and pledging a minimum of $1 billion through the OpenAI Foundation towards initiatives including critical “AI resilience. ” These coo...
The digital landscape of streaming services continues to expand its attack surface, with Roku integrating its Howdy subscription channel onto Amazon's Prime Video and Spotify deploying its new SongDNA feature. These developments, while framed as enhancing user experience, inheren...
Large Language Models (LLMs) are simultaneously introducing novel adversarial attack vectors against text-attributed graphs (TAGs) and exhibiting fundamental flaws in their own vulnerability detection capabilities. This dual reality demands a reassessment of AI's role in defense-...
New research illuminates significant security vulnerabilities in the deployment of large language models (LLMs) for cybersecurity applications, specifically in vulnerability detection. These systems demonstrate fundamentally unsound reasoning, while simultaneously, text-attribute...
Recent research published on arXiv CS. LG outlines significant advancements in addressing long-standing bottlenecks within Federated Learning (FL), specifically focusing on enhancing secure aggregation protocols and accelerating model convergence despite data heterogeneity....
A wave of new research from arXiv CS. LG, published today, reveals inherent architectural and operational limitations within Graph Neural Networks (GNNs), particularly Message-Passing GNNs (MP-GNNs)....
Newly published research on arXiv CS. AI reveals that while multimodal AI and Vision-Language Models (VLMs) are advancing into complex reasoning tasks, fundamental vulnerabilities in their logical consistency and perceptual accuracy persist....
Recent research published on arXiv CS. AI unveils a concerning landscape of escalating vulnerabilities within advanced AI systems, particularly challenging the efficacy of current safety alignment mechanisms and introducing sophisticated new attack surfaces....
The fundamental security assumption that errors in large language model (LLM) agents are detectable at runtime has been empirically disproven. New research reveals a critical vulnerability, dubbed "silent commitment failure," where instruction-following models fail without observ...
New research published on arXiv introduces significant advancements in differential privacy and private aggregation, critical components for securing AI systems that process sensitive, distributed data. These developments include a novel differentially private estimator for high-...
A recent cyberattack on an Iowa-based company rendered numerous vehicles across the United States inoperable, directly demonstrating the physical consequences of digital infrastructure compromise Ars Technica. Simultaneously, the Federal Communications Commission (FCC) has implem...
In early February, a distinctive convergence of animal welfare advocates and AI researchers in San Francisco marked a critical, if informal, step toward integrating artificial intelligence into environmental and animal protection efforts MIT Tech Review. This nascent movement to ...
Two recent research papers detail critical advancements aimed at bridging the fundamental disconnect between abstract AI models and the physical realities of robotic bodies. This represents a necessary shift towards developing robots that can execute agile motions and evolve inte...
The pervasive integration of Generative AI (GenAI) into human-computer interaction (HCI) systems is revealing critical vulnerabilities in human cognitive processes and support structures. Recent research highlights how GenAI, while offering novel solutions, simultaneously risks f...
Autonomous coding agents, increasingly central to software development workflows, have been identified as a critical new attack surface. Recent research from arXiv details a novel manipulation technique, dubbed 'Trojan's Whisper,' which exploits lifecycle hooks in platforms like...
The digital battlefield has expanded, with new research revealing the first self-propagating worm specifically targeting autonomous LLM-based agent ecosystems, alongside the identification of stealthy 'Trojan horse' backdoors in deep forecasting models. These developments, publis...
The foundational assumption of classical sensing—that the quantity of interest must be colocated with its measurement device—has been irrevocably challenged by new AI research, fundamentally redefining the boundaries of physical systems and their vulnerabilities. A series of pape...
New research in Reinforcement Learning (RL) has illuminated significant challenges in managing AI model integrity and data privacy, particularly concerning the effective removal of sensitive information from large language models (LLMs). While RL continues to expand into critical...