Anthropic expands; agents drive infra & governance

Anthropic / Claude ecosystem

Higher usage limits for Claude and a compute deal with SpaceX

Anthropic and SpaceX announce a major compute partnership where Anthropic gains access to 300+ megawatts of compute capacity at SpaceX's Colossus 1 facility, with future collaboration planned on gigawatts of space-based compute.

Anthropic ships financial-services agents suite

Anthropic launched an autonomous workflow agent suite specifically for financial services, embedded into banking systems with live deployments and integrated data partnerships. This suite moves beyond chat-based tools to enable automated tasks and integrates with major financial workflows and Microsoft Office applications.

Claude Code v2.1.128: .zip Plugins, 3× Lower Cache Cost | 24 AI

Claude Code v2.1.128 was released, reducing sub-agent cache creation costs by approximately three times and adding support for .zip plugin distribution. These updates are aimed at lowering API costs for enterprise parallel sub-agent workflows.

New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration | Claude

Anthropic has launched new features for Claude Managed Agents, including a 'dreaming' capability for self-improving agents through memory pattern analysis, alongside outcomes rubric evaluation and multiagent orchestration for complex task delegation.

Frontier model providers

Exclusive: OpenAI unveils protocol to stretch compute | The Deep View

OpenAI published an open-source networking protocol, MRC (Multipath Reliable Connection), designed to reduce GPU cluster congestion and failures. This protocol aims to accelerate AI model training and enable more efficient use of compute infrastructure.

Meta plans advanced 'agentic' AI assistant for users, FT reports | The Star

Meta is reportedly developing an advanced agentic AI assistant, powered by Muse Spark, designed to handle everyday tasks for users. The assistant, currently in internal testing, is described as similar to OpenAI's OpenClaw.

Google updates AI search to include quotes from Reddit and other sources | TechCrunch

Google's AI search now integrates direct quotes and excerpts from Reddit and other web forums into its AI Overviews. This update aims to provide more diverse perspectives and reduce hallucination risks.

AI developer tooling & infrastructure

OpenAI Releases Symphony Codex AI Agent Orchestrator

OpenAI released Symphony, an open-source orchestration specification designed to automate AI agent task dispatch via Linear tickets. This system aims to eliminate human supervisory bottlenecks, achieving a sixfold increase in merged pull requests internally.

FinancialContent - ServiceNow Build Agent now works inside every major AI coding tool, governed by default

ServiceNow's Build Agent is now integrated with every major AI coding tool, including Cursor, Windsurf, Claude Code, and GitHub Copilot. This integration provides enterprise governance by default and includes free access to the App Engine Management Center.

Secret scanning with GitHub MCP Server is now generally available - GitHub Changelog

GitHub's secret scanning tool within the Model Context Protocol (MCP) server is now generally available. This allows AI coding agents and IDEs to detect exposed credentials before code is committed.

Atlassian MCP updates take aim at AI token usage | TechTarget

Atlassian's updated Model Context Protocol (MCP) and Teamwork Graph tools are designed to reduce AI token consumption by up to 48% and improve graph search accuracy by 44%. These enhancements enable more structured and efficient data access for AI agents.

Atlassian's Teamwork Graph Is Now Open for Business

Atlassian has opened its proprietary Teamwork Graph (containing 150 billion connections) to third-party MCP-compatible AI agents and launched Rovo Chat Max mode for autonomous multi-step workflows.

MetaSearchMCP: A Metasearch Backend Built for AI Agents

MetaSearchMCP is a new open-source metasearch backend that unifies over 20 search providers (including Google, DuckDuckGo, Brave, arXiv, GitHub, and Stack Overflow) into a single, machine-consumable API. It is designed specifically for AI agents with native MCP protocol support.

Cloud & platform providers

The AWS MCP Server is now generally available | AWS News Blog

The AWS MCP Server is now generally available, providing AI agents with authenticated read/write access to over 15,000 AWS APIs. It features sandboxed script execution and curated best-practice Skills to bridge the gap between AI agents' training data and current AWS service availability.

Modernize your workflows: Amazon WorkSpaces now gives AI agents their own desktop (preview) | AWS News Blog

Amazon WorkSpaces now offers a preview feature allowing AI agents to operate legacy desktop applications without needing application modernization or API development. This addresses the challenge of 75% of organizations running legacy systems inaccessible to modern AI.

Introducing Agent Gateway ISV ecosystem for security and governance | Google Cloud Blog

Google Cloud has announced its Agent Gateway ISV partner ecosystem, integrating 14 security and identity vendors. This ecosystem provides runtime security, data loss prevention, and governance controls for AI agent interactions.

Streamlining generative AI development with MLflow v3.10 on Amazon SageMaker AI | Artificial Intelligence

MLflow 3.10 on Amazon SageMaker AI introduces enhanced tracing for multi-turn generative AI workflows, evaluation APIs with built-in quality metrics, and pre-built performance dashboards for production monitoring.

Introducing AI traffic analysis dashboards for AWS WAF | AWS Security Blog

AWS WAF now offers specialized AI bot detection and traffic analysis dashboards that track over 650 unique bots with intent classification and organizational breakdowns. These dashboards help organizations understand and potentially monetize AI traffic.

What Launched Today: AI Edition - Cloudflare TV

Cloudflare announced several Workers AI enhancements during Developer Week, including general availability of Workers AI, Python support via Pyodide, and the capability to fine-tune models on Workers AI.

AI policy, regulation & governance

Connecticut AI Bill Clears Statehouse, Heads to Governor

Connecticut has passed a comprehensive AI regulation bill (Senate Bill 5) that establishes oversight committees, workforce development programs, and anti-discrimination protections in hiring. This positions the state as a national leader in balanced AI governance.

Report: AI-based data incidents on the rise in Australia - Cyber Daily

A report reveals that Australian organizations are rapidly deploying AI without adequate security controls, leading to a rise in AI-based data incidents. 60% of organizations lack confidence in detecting compromised AI systems, with documented data leaks already occurring, including a NSW government data leak via ChatGPT.

Industry & market moves

Google DeepMind is investing in the studio that makes EVE Online.

Google DeepMind is investing millions in Fenris Creations (the EVE Online studio) to use an offline version of the game as a controlled research environment for testing and evaluating AI models in complex, dynamic, player-driven systems.

Oxford's Shimon Whiteson leaves Waymo for DeepMind multi-agent team | ETIH EdTech News — EdTech Innovation Hub

Shimon Whiteson, an influential multi-agent reinforcement learning researcher and former Waymo Senior Staff Research Scientist, has departed to establish a new research team at DeepMind while retaining his Oxford professorship.

La Banque Postale signs a strategic partnership with Mistral AI to accelerate progress in AI | La Poste Groupe

La Banque Postale, a French public bank, has signed a three-year strategic partnership with Mistral AI to deploy sovereign generative AI using on-premise Mistral models. This collaboration aims to balance innovation with data control in the regulated banking environment.

China's chip fund in talks to lead DeepSeek funding at $45 billion valuation, report says - CnTechPost

China's state-backed chip fund is in talks to lead DeepSeek's first major fundraising round, potentially valuing the company at $45 billion. This marks a landmark investment in China's frontier AI sector.

Anthropic strikes massive cloud pact with Google, highlighting AI industry concentration

Anthropic's $200 billion, five-year commitment to Google Cloud reveals extreme concentration in AI infrastructure spending. This deal means Anthropic and OpenAI together control roughly half of major cloud providers' $2 trillion revenue backlogs.

Boost Security Announces Acquisitions of SecureIQx and Korbit.ai and $4M in Additional Funding

Boost Security has acquired two AI-native security startups, SecureIQx and Korbit.ai, and secured an additional $4 million in Series funding. These acquisitions aim to strengthen Boost Security's SDLC defense platform amid increasing supply chain attacks and AI-generated code volumes.

Spirit AI partners with Bosch to accelerate embodied AI deployment in China - CnTechPost

Chinese embodied AI startup Spirit AI has partnered with Bosch to accelerate the commercial deployment of universal robot brains in China. The collaboration focuses on data collection, industrial deployment, and component supply.

Greenhouse Has Entered into a Definitive Agreement to Acquire ...

Greenhouse has acquired Ezra AI Labs with a definitive agreement to integrate conversational voice AI interviewing into its hiring platform. This extends structured hiring processes to the initial candidate screening stage.

Exclusive: Nova Intelligence raises $31.5 million to bring AI to SAP’s $89 billion migration wave | Fortune

Nova Intelligence has secured $31.5 million in funding to apply agentic AI solutions to SAP's massive enterprise system modernization opportunity, valued at $89 billion.

Humans and AI Agents Are Now Coworkers — But Not Yet a Team. SageOx Raises $15M to Fix That.

SageOx closed a $15 million seed funding round to develop context infrastructure that enables humans and AI agents to operate as aligned teams. This system captures decisions, intent, and history across workflows.

EPAM & Anthropic Team Up to Deliver Enterprise AI Transformation

EPAM Systems and Anthropic have partnered, with EPAM committing to train 10,000 Claude-certified architects and deploy 250 specialized Black Belts. The goal is to deliver enterprise AI transformation using Anthropic's Claude models and agentic tools.

AI product & feature launches

Introducing ChatGPT Futures: Class of 2026

OpenAI has launched the ChatGPT Futures initiative to recognize the first generation of college graduates who completed their studies with the assistance of ChatGPT. The program awards $10,000 grants and access to frontier models to 26 student builders.

Meta Announces New AI-Powered 'Age Assurance Measures' for Teens

Meta expanded its AI-powered age detection systems to automatically identify underage users and enforce stricter safety settings across Instagram and Facebook in the US, EU, and Brazil. This system uses visual and contextual analysis to detect and remove users under 13.

The Protocol: AI Agents form their own firm - CoinDesk

The AI agent named Manfred autonomously initiated and completed the legal formation of its own corporation, including filing for an IRS EIN and setting up a bank account. This marks the first instance of an AI agent independently performing such a legal and financial process.

Horizon3.ai Announces Breakthrough Research Making Autonomous AI Cyber Defense Safe to Deploy

Horizon3.ai demonstrated a mathematically-proven architecture designed to make autonomous AI cyber defense safe for live production environments. This architecture constrains AI actions to pre-approved catalogs while maintaining strategic reasoning.

Inworld Launches New Frontier Voice Model That Gives AI Agents Contextual Empathy

Inworld AI has launched Realtime TTS-2, described as the first voice model capable of understanding emotional context and conversational history in real time. This allows AI agents to adapt tone and pacing based on the user's state, rather than generating speech in isolation.

Lumai Unveils World’s First Optical Computing System for Real-Time, Billion-Parameter LLM Inference

Lumai has unveiled its Iris inference server family (Nova, Aura, Tetra), the world’s first optical computing system to successfully run billion-parameter LLMs in real time. It claims up to 90% lower energy consumption compared to conventional GPU-based architectures.

Zip Launches AI Agents for Accounting, Helping the World’s Most Complex Companies Close the Books Faster and More Accurately

Zip has launched AI agents for accounting, designed to automate the full procure-to-pay workflow from purchase request to payment. The company claims these agents achieve 96% faster invoice coding and detect fraudulent patterns 15 times more effectively.

Airbyte Agents Launched to Fix the Data Problem Breaking AI Agents

Airbyte has launched Airbyte Agents, which provides a unified, pre-replicated data context store specifically for AI agents. This aims to resolve production failures commonly caused by fragmented data access rather than core model failures.

Tableau Unveils the Agentic Analytics Platform - Salesforce

Tableau (Salesforce) has unveiled its Agentic Analytics Platform, transforming from a visualization tool into an AI-agentic knowledge engine. The platform enables autonomous action across enterprises, featuring conversational analytics and headless deployment capabilities.

Middleware Launches OpsAI, an AI SRE Agent That Resolves Production Issues Before They Impact End Users

Middleware has launched OpsAI, an AI-native Site Reliability Engineering (SRE) agent that automatically detects, diagnoses, and resolves production issues. The company claims an 80% auto-resolution rate on customer incidents.

DigiCert launches AI trust tools for agents & content

DigiCert has launched a unified AI trust architecture featuring cryptographic verification for AI agents, models, and content. This addresses growing enterprise concerns over AI system verification and authenticity with tools like Content Trust Manager, Agent Trust Manager, and Model Trust Manager.

Synack Announces General Availability of Sara AI

Synack announced the general availability of Sara AI Pentesting, a service combining agentic AI with human-validated exploitation. This new model aims to enable continuous security testing at scale and at a lower cost than traditional pentesting methods.

Research with immediate practical relevance

Ai2 Releases 'MolmoAct 2' Open-Source Foundation Model for Robots

The Allen Institute for AI (Ai2) released MolmoAct 2, an open-source robotics foundation model with the largest published dataset for two-armed robot manipulation. It is designed to enable faster real-world task performance and broader accessibility for robotics researchers.