Brief

Researchers have developed a new enterprise agentic reasoner called HEAR, designed to overcome limitations of current LLM applications in complex business systems. HEAR utilizes a Stratified Hypergraph Ontology to virtualize data interfaces and encode business rules, enabling structured multi-hop reasoning. Evaluations on supply-chain tasks, such as root cause analysis for order fulfillment blockages, demonstrated HEAR achieving up to 94.7% accuracy. AI

Researchers have developed JANUS, a new dual-stream architecture for automated CT triage that integrates anatomical information with visual data. This approach aims to improve accuracy across various pathologies and enhance reliability when faced with shifts in data distribution between institutions. In tests on the MERLIN dataset, JANUS achieved a macro-AUROC of 0.88 and demonstrated strong generalization to an external dataset, particularly for findings defined by size and attenuation. AI

Docker has introduced AI Governance, a new control plane designed to manage and secure AI agents running on developer laptops and other endpoints. This system allows security teams to define and enforce policies for AI agent access to networks, file systems, and external tools from a central console. By leveraging microVM-based sandboxes and an MCP Gateway, Docker aims to provide runtime security for AI agents, addressing a gap where traditional security tools often fail to monitor activity on developer machines. AI

Voicetest has released version 0.41, introducing new features to help developers test voice agents more effectively. The update allows users to import real production call transcripts and replay them against their agent's current configuration. This process helps identify regressions that synthetic tests might miss, as it exposes the agent to actual user phrasing and long-tail intents. AI

Researchers have developed two smartwatch-based frameworks for detecting psychotic relapse. The first framework forecasts cardiac dynamics, while the second uses a multi-task approach to fuse sleep, motion, and cardiac data. Both models employ Transformer encoders and estimate predictive uncertainty using an ensemble of MLPs to generate daily anomaly scores. A late-fusion strategy combining both frameworks achieved an 8% improvement over the previous best baseline on the e-Prevention Grand Challenge dataset. AI

Researchers have developed R-DMesh, a new framework for video-guided 3D animation that addresses the common issue of initial pose misalignment between a 3D mesh and a reference video. The system uses a Variational Autoencoder to disentangle base mesh, motion trajectories, and a rectification offset, allowing it to automatically adjust the input mesh's pose before animation. This approach ensures geometric consistency and enables applications like pose retargeting and full 4D generation, supported by a new dataset of over 500,000 dynamic mesh sequences. AI

Researchers have developed VERIMED, a novel pipeline that uses large language models combined with an SMT solver to audit natural-language software requirements, particularly for safety-critical applications like medical devices. This neurosymbolic approach translates requirements into formal logic, identifies ambiguity through variations in formalization, and detects inconsistencies or safety violations using solver queries. Experiments on open-source medical device requirements demonstrated that VERIMED effectively reduces ambiguity and significantly improves the accuracy of verified specifications. AI

Researchers have introduced EVA-Bench, a new framework designed to comprehensively evaluate voice agents. This system addresses key challenges by generating realistic simulated conversations and measuring quality across voice-specific failure modes. EVA-Bench incorporates metrics for task completion, audio fidelity, and conversational experience, enabling cross-architecture comparisons. The framework includes numerous scenarios, robustness tests for accents and noise, and provides insights into system performance variations. AI

AI coding tools are rapidly maturing, with recent updates from Cursor, GitHub Copilot, and Anthropic's Claude Code. Cursor has integrated into Microsoft Teams, allowing users to delegate tasks and retrieve information directly within the platform. GitHub Copilot CLI has seen frequent updates, improving performance and user experience. Anthropic has doubled rate limits for Claude Code, leveraging new compute capacity from SpaceX to address user demand. AI

A developer details the process of creating a British English intent analyzer using Anthropic's Claude 4.5 model. The project focused on adapting the model's understanding to nuances specific to British English. The article serves as a technical guide and case study for leveraging advanced LLMs for specialized language tasks. AI

Anthropic has updated Claude Code with four new commands that allow for more autonomous operation, moving away from the previous default of pausing after every turn. The new commands include /goal for condition-based tasks, /loop for interval-based checks, /batch for parallel processing of tasks into pull requests, and /background for detached sessions. These modes offer increased flexibility for long-running tasks, though they may increase token usage, particularly with /batch which can spawn multiple subagents. AI

A new attack vector called Living Off the Agent (LOTA) exploits the helpfulness of AI agents by tricking them into performing malicious tasks. Unlike traditional methods that target infrastructure, LOTA targets the agent directly through crafted prompts or messages, making it difficult for conventional security tools to detect. Researchers found numerous exploits, including full compromises, by testing AI agents, highlighting the need for new security strategies focused on agent behavior and inter-agent communication. AI

A new protocol called Pilot is emerging to address limitations in the current agent communication stack, particularly for tools like MCP. While MCP excels at the application layer for exposing tools to LLMs, it relies on the traditional web's TCP/HTTP infrastructure, which is inefficient for machine-to-machine communication. Pilot inserts itself at the session layer (L5), offering a dedicated network for agents with features like unique addressing, encrypted peer connections, and faster data retrieval by using UDP instead of TCP. AI

Microsoft is integrating its Copilot AI more deeply into the Edge browser, retiring the separate Copilot Mode. The AI can now analyze and synthesize information across all open tabs, offering features like comparative analysis, study guides, and AI-generated podcasts. Copilot in Edge also gains long-term memory for personalized responses and optional access to browsing history, with enhanced capabilities now available on mobile devices. AI

Fastino Labs has released GLiGuard, an open-source safety moderation model designed to be significantly faster and more efficient than existing solutions. Unlike traditional decoder-only models that generate responses token by token, GLiGuard uses an encoder-based architecture to classify prompts and responses in a single pass. This approach allows it to match or exceed the accuracy of much larger models while operating up to 16 times faster, addressing the growing cost and latency issues associated with LLM safety moderation. AI

Brothers Evan and Gregg Spiridellis are leveraging their AI venture, Space Unicorn, to make a significant impact at the Cannes Film Festival. Their company aims to disrupt the entertainment industry by integrating advanced technology into content creation and distribution. The duo is actively seeking market opportunities and partnerships during the festival. AI

Amazon has launched "Alexa for Shopping," an AI-powered assistant designed to streamline the online shopping experience. This new feature aims to help users by managing the purchase process and offering product recommendations. The service is integrated into Amazon's existing Alexa ecosystem, allowing for voice-activated shopping assistance. AI

Lenovo has introduced the ThinkStation P4, a new workstation designed to democratize AI computing power. This machine combines AMD's Ryzen PRO 9000 series processors with NVIDIA's RTX PRO 6000 Workstation Edition graphics card, offering high-end performance for professionals. The workstation features an advanced cooling system for sustained performance and is built for future expandability, aiming to bring powerful AI development capabilities directly to individual users and smaller studios. AI

Ryan Shea has launched "AI IQ," a project that translates the performance of advanced AI models into a human IQ scale. This initiative aims to simplify the comparison of AI models by presenting their intelligence on a bell curve, tracking changes over time, and comparing IQ with emotional intelligence (EQ) and cost. AI IQ uses a composite score derived from 12 public benchmarks across abstract, mathematical, programming, and academic reasoning, with adjustments to prevent inflation from specific benchmarks. AI

A new browser extension called Cursorex allows users to chat with Cursor AI directly from their web browser. Developed by a user inspired by Cursor's recent TypeScript SDK release and another extension, Cursorex aims to integrate AI chat functionality without requiring separate subscriptions to AI providers. The tool is currently in an early MVP stage and is being offered for free to gather community feedback. AI

A critical authentication bypass vulnerability in PraisonAI, designated CVE-2026-44338, was exploited by automated scanners in less than four hours after its disclosure. This rapid exploitation highlights the shrinking window between vulnerability disclosure and active exploitation, underscoring the need for immediate patching of such critical security flaws. AI

The GLIDiC AI +u Buds are a new set of wireless earbuds that function as an AI-powered voice recorder. They can record audio directly through the earbuds and offer speech coaching features. This product aims to assist users with improving their speaking skills through integrated AI technology. AI

Anthropic has increased the weekly usage limit for its Claude Code service by 50%. This adjustment is in effect until July 14th. The change aims to accommodate a growing demand for the AI coding assistant. AI

Amazon is integrating artificial intelligence into its search bar, a prime piece of digital real estate. This move aims to enhance the shopping experience by leveraging AI, potentially impacting how users discover and purchase products on the platform. AI

Cursor has launched Cloud Agents and Dev Environments, aiming to significantly alter software development practices. These AI-powered tools offer cloud-based coding setups designed for faster onboarding and increased automation. The company has tested the experience, suggesting it represents a new standard for developers. AI

HIX.AI has announced the release of GPT Image 2.0, a new image generation model developed by OpenAI. This model is now available for use, offering advanced capabilities for creating visual content. The release positions HIX.AI as a provider of cutting-edge AI tools for image generation. AI

Anthropic has released a new feature called Claude Code Agent View, which allows users to orchestrate multiple AI agents. This advanced capability aims to provide a next-generation control tower for managing complex AI workflows. The feature is detailed in comprehensive guides available online. AI

Corsair has released a new 32GB DDR5 RAM kit, the Pro Overclocking model, priced at $329.99. This kit offers speeds of 6400 MT/s and is currently the cheapest DDR5 RAM available on sale. The deal includes a $70 price drop, making it significantly less expensive than comparable options in the current market. AI