Agents drive innovation; EU rolls back AI rules

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

No significant new developments.

AI developer tooling & infrastructure

Cursor 3.3 Introduces Context Telemetry to Solve Agent Reasoning Issues

Cursor 3.3 adds context telemetry to show how AI agents consume token budgets across rules, skills, MCPs, and subagents, helping developers diagnose reasoning degradation.

Memori Labs Releases New Agent-Native Memory Infrastructure, Automatically Creating Structured Memory from Agent Trace

Memori Labs releases first agent-native memory system that structures memory from agent execution traces (tool calls, decisions, workflow steps) rather than conversation alone, achieving 81.95% accuracy on LoCoMo benchmark with 95% inference cost savings.

Cloud & platform providers

AWS Skill Builder introduces Lab Maker, an AI-powered feature that generates personalized, step-by-step guided labs for AWS services in natural language.

WNBA Inside the Game, powered by AWS: AI Meets the Next Era of Women’s Basketball | AWS for M&E Blog

AWS brings AI-powered Gravity analytic to WNBA for the first time, quantifying defensive pressure and player impact on spacing in live broadcasts and digital channels.

Closing the Loop with Amazon Bio Discovery’s Integrated Lab Partners | AWS for Industries

Amazon Bio Discovery integrates computational drug design with live lab-in-the-loop workflows through a network of CRO partners, removing friction from iterative antibody discovery.

AI policy, regulation & governance

Anthropic Skill scanners passed every check. The malicious code rode in on a test file. | VentureBeat

Malicious code in bundled test files bypasses all three major Anthropic Skill scanners because they monitor the agent execution surface but not the developer toolchain layer where test files execute with full local permissions.

EU clinches deal to roll back AI restrictions – POLITICO

EU clinches deal to delay high-risk AI restrictions by 14+ months and largely exempts industrial AI applications, marking the first significant rollback of EU digital rules amid global competition pressure.

Industry & market moves

Musk Dissolves xAI After Anthropic Deal, Future AI Products to Launch Under SpaceXAI | Artificial Intelligence News - News9live

Musk dissolves xAI as standalone entity and merges AI operations under new SpaceXAI banner within SpaceX, consolidating rocket, satellite, and AI computing infrastructure.

Renesas Completes Acquisition of Irida Labs to Expand Vision AI Software Capabilities and Accelerates System-Level Vision Solutions

Renesas completes acquisition of Irida Labs to integrate embedded Vision AI software into its edge AI processing portfolio and newly launched Renesas 365 platform.

IREN Announces Acquisition of Mirantis to Strengthen AI Cloud Delivery Capabilities - StorageNewsletter

IREN acquires Mirantis to strengthen AI cloud delivery with Kubernetes orchestration and enterprise support capabilities for at-scale GPU infrastructure deployments.

Roche Expands Digital Pathology Strategy with Acquisition of AI Diagnostics Company PathAI | EuropaWire

Roche acquires PathAI for USD 750M upfront to integrate AI-powered digital pathology and diagnostics into its oncology and precision medicine portfolio

Immunai and AstraZeneca Expand AI-Driven Biomarker Project | The AI Journal

Immunai's expansion with AstraZeneca signals successful embedding of single-cell AI immune mapping into major pharma R&D workflows for oncology and IBD drug development.

NVIDIA and IREN Announce Strategic Partnership to Accelerate ...

NVIDIA and IREN announced a strategic partnership to deploy up to 5 gigawatts of AI infrastructure globally, combining NVIDIA's DSX architecture with IREN's power and data center expertise, with IREN issuing NVIDIA a $2.1B share purchase right.

SpaceX has a $55 billion plan to build AI chips in Texas | The Verge

SpaceX's $55 billion Terafab chip manufacturing plant in Texas represents a major expansion into AI chip production, with Intel as a design and fabrication partner.

Scale AI wins $500m Pentagon contract, five times its previous deal

Scale AI's $500M Pentagon contract represents a five-fold expansion of its defense relationship and validates data-labelling as a separate procurement category from cloud infrastructure.

AI product & feature launches

Meta announces AI support for SMBs and rural companies | Social Media Today

Meta launches two initiatives to help U.S. SMBs and rural businesses adopt AI tools through an incubator partnership with gener8tor and a Community Accelerator training program.

Greenhouse Launches MCP, Giving Hiring Teams a Governed Way to Connect AI Tools to Greenhouse

Greenhouse launches a governed API layer (MCP) enabling AI agents to safely connect to hiring platform with permission controls and audit trails.

Pluralsight Launches Cloud Ready to Help Enterprises Build AI-Ready Cloud Teams

Pluralsight launches an integrated enterprise cloud upskilling program combining skill assessments, guided learning, instructor-led training, and hands-on labs to help organizations close cloud skills gaps and prepare for AI at scale.

Genesis AI introduces GENE-26.5 robotic model - Engineering

Genesis AI's GENE-26.5 is the first robotic brain to achieve human-level physical manipulation capabilities by closing the embodiment gap through a human-hand-matching robotic hand and low-cost data collection glove.

Spotify adding deeper AI agent integration with 'Personal Podcasts'

Spotify is enabling AI agents to generate and save personalized podcast content directly to user libraries, integrating calendar, inbox, and feed data.

MongoDB Makes Enterprise AI Production Ready

MongoDB delivers native embeddings generation, persistent agent memory, and real-time operational data in a unified platform to enable enterprises to run AI agents in production without stitching together disparate systems.

Research with immediate practical relevance

Anthropic Introduces Natural Language Autoencoders That Convert Claude's Internal Activations Directly into Human-Readable Text Explanations - MarkTechPost

Anthropic's Natural Language Autoencoders convert a model's internal activations into human-readable text explanations, enabling direct interpretation of what Claude 'thinks' without requiring trained researchers to manually decode complex outputs.