AI

Onchain RWA Tops $10 Billion and Tokenized Stocks Hit $1B as Institutional Adoption Grows

NVIDIA AI Unveils ProRL Agent: A Decoupled Rollout-as-a-Service Infrastructure for Reinforcement Learning of Multi-Turn LLM Agents at Scale

Anthropic’s Claude Mythos AI Model Exposed in Major Data Breach

Anthropic’s massive 'Claude Mythos' leak sends software names — and crypto — sharply lower

SoftBank Obtains Massive $40B Loan for Additional OpenAI Investment

OpenAI Sits at the Center of a $1.4 Trillion Capital Loop, Morgan Stanley Warns

Morgan Stanley warns that OpenAI sits at the center of a massive $1.4 trillion capital loop, with infrastructure commitments far exceeding its $13 billion annual revenue. Major tech companies like Microsoft, Amazon, Oracle, and Nvidia are both funding OpenAI and receiving its spending commitments back, creating potential hidden leverage risks for investors.

ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence

Researchers introduce ARC-AGI-3, a new benchmark for testing agentic AI systems that focuses on fluid adaptive intelligence without relying on language or external knowledge. While humans can solve 100% of the benchmark's abstract reasoning tasks, current frontier AI systems score below 1% as of March 2026.

When Is Collective Intelligence a Lottery? Multi-Agent Scaling Laws for Memetic Drift in LLMs

Researchers introduce Quantized Simplex Gossip (QSG) model to explain how multi-agent LLM systems reach consensus through 'memetic drift' - where arbitrary choices compound into collective agreement. The study reveals scaling laws for when collective intelligence operates like a lottery versus amplifying weak biases, providing a framework for understanding AI system behavior in consequential decision-making.

DRIFT: Dynamic Rule-Based Defense with Injection Isolation for Securing LLM Agents

Researchers introduce DRIFT, a new security framework designed to protect AI agents from prompt injection attacks through dynamic rule enforcement and memory isolation. The system uses a three-component approach with a Secure Planner, Dynamic Validator, and Injection Isolator to maintain security while preserving functionality across diverse AI models.

Sparse Visual Thought Circuits in Vision-Language Models

Research reveals that sparse autoencoder (SAE) features in vision-language models often fail to compose modularly for reasoning tasks. The study finds that combining task-selective feature sets frequently causes output drift and accuracy degradation, challenging assumptions used in AI model steering methods.

Beyond Content Safety: Real-Time Monitoring for Reasoning Vulnerabilities in Large Language Models

Researchers have identified a new category of AI safety called 'reasoning safety' that focuses on protecting the logical consistency and integrity of LLM reasoning processes. They developed a real-time monitoring system that can detect unsafe reasoning behaviors with over 84% accuracy, addressing vulnerabilities beyond traditional content safety measures.

Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia

A research study analyzing Google's AI Overviews feature found it reduces Wikipedia traffic by approximately 15% through causal analysis of 161,382 matched articles. The impact varies by content type, with Culture articles experiencing larger traffic declines than STEM topics, suggesting AI summaries substitute for clicks when brief answers satisfy user queries.

SWAA: Sliding Window Attention Adaptation for Efficient and Quality Preserving Long Context Processing

Researchers propose SWAA (Sliding Window Attention Adaptation), a toolkit that enables efficient long-context processing in large language models by adapting full attention models to sliding window attention without expensive retraining. The solution achieves 30-100% speedups for long context inference while maintaining acceptable performance quality through four core strategies that address training-inference mismatches.

Epistemic Bias Injection: Biasing LLMs via Selective Context Retrieval

Researchers have identified a new attack vector called Epistemic Bias Injection (EBI) that manipulates AI language models by injecting factually correct but biased content into retrieval-augmented generation databases. The attack steers model outputs toward specific viewpoints while evading traditional detection methods, though a new defense mechanism called BiasDef shows promise in mitigating these threats.

Ming-Flash-Omni: A Sparse, Unified Architecture for Multimodal Perception and Generation

Ming-Flash-Omni is a new 100 billion parameter multimodal AI model with Mixture-of-Experts architecture that uses only 6.1 billion active parameters per token. The model demonstrates unified capabilities across vision, speech, and language tasks, achieving performance comparable to Gemini 2.5 Pro on vision-language benchmarks.

DiffuGuard: How Intrinsic Safety is Lost and Found in Diffusion Large Language Models

Researchers identified critical security vulnerabilities in Diffusion Large Language Models (dLLMs) that differ from traditional autoregressive LLMs, stemming from their iterative generation process. They developed DiffuGuard, a training-free defense framework that reduces jailbreak attack success rates from 47.9% to 14.7% while maintaining model performance.

The LLM Bottleneck: Why Open-Source Vision LLMs Struggle with Hierarchical Visual Recognition

Research reveals that open-source large language models (LLMs) lack hierarchical knowledge of visual taxonomies, creating a bottleneck for vision LLMs in hierarchical visual recognition tasks. The study used one million visual question answering tasks across six taxonomies to demonstrate this limitation, finding that even fine-tuning cannot overcome the underlying LLM knowledge gaps.

LLM4AD: Large Language Models for Autonomous Driving -- Concept, Review, Benchmark, Experiments, and Future Trends

Researchers have published a comprehensive review of Large Language Models for Autonomous Driving (LLM4AD), introducing new benchmarks and conducting real-world experiments on autonomous vehicle platforms. The paper explores how LLMs can enhance perception, decision-making, and motion control in self-driving cars, while identifying key challenges including latency, security, and safety concerns.

LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution Shifts

Researchers have identified a new vulnerability in large language models called 'natural distribution shifts' where seemingly benign prompts can bypass safety mechanisms to reveal harmful content. They developed ActorBreaker, a novel attack method that uses multi-turn prompts to gradually expose unsafe content, and proposed expanding safety training to address this vulnerability.

AD-CARE: A Guideline-grounded, Modality-agnostic LLM Agent for Real-world Alzheimer's Disease Diagnosis with Multi-cohort Assessment, Fairness Analysis, and Reader Study

Researchers developed AD-CARE, an AI agent that uses large language models to diagnose Alzheimer's disease from incomplete medical data across multiple modalities. The system achieved 84.9% diagnostic accuracy across 10,303 cases and improved physician decision-making speed and accuracy in clinical studies.

GlowQ: Group-Shared LOw-Rank Approximation for Quantized LLMs

Researchers propose GlowQ, a new quantization technique for large language models that reduces memory overhead and latency while maintaining accuracy. The method uses group-shared low-rank approximation to optimize deployment of quantized LLMs, showing significant performance improvements over existing approaches.

How Pruning Reshapes Features: Sparse Autoencoder Analysis of Weight-Pruned Language Models

Researchers conducted the first systematic study of how weight pruning affects language model representations using Sparse Autoencoders across multiple models and pruning methods. The study reveals that rare features survive pruning better than common ones, suggesting pruning acts as implicit feature selection that preserves specialized capabilities while removing generic features.

CRAFT: Grounded Multi-Agent Coordination Under Partial Information

Researchers introduce CRAFT, a multi-agent benchmark that evaluates how well large language models coordinate through natural language communication under partial information constraints. The study finds that stronger reasoning abilities don't reliably translate to better coordination, with smaller open-weight models often matching or outperforming frontier systems in collaborative tasks.

A Wireless World Model for AI-Native 6G Networks

Researchers introduce the Wireless World Model (WWM), a multi-modal AI framework for 6G networks that predicts wireless channel evolution by understanding electromagnetic wave propagation through 3D geometry. The model demonstrates superior performance across five downstream tasks and real-world measurements, outperforming existing foundation models.

Shape and Substance: Dual-Layer Side-Channel Attacks on Local Vision-Language Models

Researchers discovered significant privacy vulnerabilities in local Vision-Language Models that use Dynamic High-Resolution preprocessing. The dual-layer attack framework can exploit execution-time variations and cache patterns to infer sensitive information about processed images, even when models run locally for privacy.