Agents drive infra, regulation; EU transparency begins

Anthropic / Claude ecosystem

Anthropic Upgrades Claude Voice Mode With Opus and Sonnet Support

Anthropic has upgraded its Claude voice mode to support its most capable models, Opus and Sonnet, enabling advanced reasoning and tool integration during spoken conversations.

Anthropic Launches 4 Role-Based Certifications And Partners With Pearson On Verified Assessment

Anthropic expanded its Claude Certification Program to include four role-based credentials and partnered with Pearson for verified, proctored assessment delivery, with major consulting firms committing to certify tens of thousands of employees.

Frontier model providers

Meta Muse Spark 1.1 reaches consumers through Meta AI

Meta Muse Spark 1.1, featuring long-context (1M token) support and multi-agent capabilities, is now integrated into Meta AI chatbot across web, mobile, and an upcoming WhatsApp rollout.

AI developer tooling & infrastructure

T3 Code Merges 76 PRs Adding Auto Approvals, Sidebar V2, And Opus 5

T3 Code shipped 76 merged PRs in one week, adding an auto-approval mode, a Sidebar v2 beta UI redesign, and Opus 5 model support to its open-source IDE.

JAMS Launches AI for Enterprise Job Scheduling: JAX and JAMS MCP

JAMS launched JAX (AI agent) and JAMS MCP (Model Context Protocol connector) to enable IT teams to query and manage job scheduling in plain language while keeping operational data onshore on customer-chosen models.

Kakao's PlayMCP Draws 80,000 Daily Calls from AI Agents

Kakao's PlayMCP platform for AI agent integration is now live with external AI services like ChatGPT, Claude, and OpenClaw, processing 80,000 daily calls and hosting 461 registered MCP servers.

System1 Launches Dogpile Fetch, a New Search MCP Server From the Internet’s Original Metasearch Engine, Giving AI Agents a Wider View of the Web | SST Stock News - BANKBIT

System1 released Dogpile Fetch, a new search MCP Server that integrates the original metasearch engine with AI agent capabilities, offering AI agents a wider view of web information.

Meituan LongCat-2.0 Open Source: 1.6T Model for Agentic Coding

Meituan has open-sourced LongCat-2.0, a 1.6 trillion-parameter model with sparse attention and N-gram Embedding specifically designed for agentic coding, including optimized inference code for domestic GPU hardware.

Cloud & platform providers

AWS Launches aws-bench Open-Source Benchmark for AI Agent Evaluation

AWS released aws-bench, an open-source benchmark designed to evaluate AI agents based on real-world AWS cloud tasks, addressing the limitations of existing synthetic benchmarks.

AI policy, regulation & governance

Lee Jae Myung meets OpenAI, Nvidia, Anthropic and Broadcom CEOs to advance Korea’s AI mega-projects

South Korean President Lee Jae Myung met with CEOs from OpenAI, Nvidia, Anthropic, and Broadcom to secure commitments for national AI mega-projects, including chip partnerships, sovereign AI development, and an AI research lab.

Google commits $40M to the Genesis Mission

Google has committed $40 million in AI tokens and cloud credits to the White House Genesis Mission, providing access to frontier AI tools and Gemini for Government to all 17 DOE National Laboratories.

Microsoft, Meta, Nvidia, OpenAI, and Palantir have a message for Washington

32 major tech companies and organizations jointly signed a letter urging US policymakers to protect open-weight AI models from restrictions, amid rising tensions over China-led model development and potential government crackdowns.

Google signs EU AI Act transparency code and expands SynthID watermarking to five AI partners including Apple and OpenAI

Google formally committed to EU transparency standards under the AI Act and expanded its SynthID watermarking technology to five major AI partners (Apple, ElevenLabs, Kakao, NVIDIA, OpenAI), positioning it as mandatory infrastructure ahead of the August 2 enforcement deadline.

Under Federal Bill, AI Companies Would Need a ‘Kill Switch’

A bipartisan federal bill in the US Congress would mandate AI companies install 'kill switches' and empower the DHS to order model shutdowns in emergency situations, with penalties up to $20M/day for non-compliance.

CPI | New EU Guidance on AI Transparency: What Should Companies Be Do…

The European Commission has published draft Guidelines interpreting AI Act transparency obligations and a Code of Practice for labelling AI-generated content, effective August 2, 2026, providing businesses with clearer compliance expectations.

EU rewrites the AI Act's clock and bans nudification tools in one stroke

The EU is delaying high-risk AI Act compliance deadlines by up to 16 months while simultaneously introducing a December 2026 ban on non-consensual intimate imagery generation tools with top-tier penalties.

Why clearer rules on using AI in hiring would be a win for bosses too

Victoria, Australia, is proposing new laws to regulate AI in hiring and workplace surveillance, including bans on emotion recognition and mandatory human oversight of algorithmic decisions.

Industry & market moves

Microsoft Weighs Chinese Kimi K3 Model for Selected Copilot Tasks

Microsoft is evaluating Moonshot AI's Kimi K3 model for selective Copilot tasks, potentially leading to cost reductions of up to $600M annually.

Prentis, new AI lab co-founded by Reid Hoffman, Mark Pincus in talks to raise $100M

A new AI lab named Prentis, co-founded by serial entrepreneurs Reid Hoffman and Mark Pincus, is reportedly in talks to raise $100 million at a $1 billion valuation to scale computer-use models for office automation.

South Korea's Naver is partnering with Nvidia to expand its sovereign AI infrastructure, tripling the size of its Gak Sejong data center and aiming for an 18x increase in capacity toward gigawatt scale, backed by $6.8 billion in external financing.

United States Industrial AI Startup Atoms Raised $1.7 billion in New Funding Round, Launched in 2026 March by Travis Kalanick (Uber Founder) with Various Businesses Merging into Atoms, Investors Include a16z, Bain Capital, Uber, Fifth Wall, Chemistry, A*, K5 Global, Abstract, SV Angel & Alpha Square Group

Industrial AI robotics startup Atoms, founded by Uber's Travis Kalanick in March 2026, has closed a $1.7 billion Series funding round with backing from prominent investors including a16z, Bain Capital, and Uber.

https://blockonomi.com/stripe-eyes-10b-acquisition-of-ai-model-marketplace-openrouter/

Stripe is reportedly in advanced negotiations to acquire AI model marketplace OpenRouter for approximately $10 billion, marking a major expansion into AI infrastructure.

Databricks and Microsoft Signal Where Enterprise AI Is ...

Databricks and Microsoft have deepened their enterprise AI collaboration, embedding Databricks' data platform and AI tools into Microsoft 365, Teams, and Power BI, and increasing Azure Cobalt infrastructure adoption.

Altana Acquires Customs Automation Startup Cervo AI

Altana has acquired customs automation startup Cervo AI in a deal worth over $100 million, aiming to automate customs paperwork and scale its supply chain compliance platform.

AI product & feature launches

FLUX 3 Launches: Black Forest Labs Enters Video, Audio, and Physical AI in One Model

Black Forest Labs has launched FLUX 3, a multimodal generative model that unifies image, video, audio, and robot action prediction within a single shared backbone architecture, with production deployment already running at Audi.

Meituan Open-Sources LongCat-Video-Avatar1.5, Outperforming Mainstream Closed Models

Meituan has open-sourced LongCat-Video-Avatar 1.5, a digital human video model that reportedly outperforms closed commercial systems like HeyGen and Kling Avatar 2.0 with 15x faster inference and improved stability.

Research with immediate practical relevance

Naukowcy: agentów kodujących AI można oszukać podrobionymi…

Researchers demonstrated 'HalluSquatting,' a new attack exploiting AI hallucinations in coding agents to inject malicious packages and achieve remote code execution across multiple platforms.

Ant Open Source releases LLaDA2.2-flash, an agent‑oriented MoE diffusion LLM with Levenshtein self‑editing

Ant Open Source has released LLaDA2.2-flash, a self-editing diffusion model with Levenshtein token editing, enabling real-time self-correction during generation and addressing model collapse in diffusion-based text generation.

BAAI Introduces AREX for Recursive Self-Improving Deep Research

The Beijing Academy of Artificial Intelligence (BAAI) introduced AREX, a framework that overcomes persistent AI errors in long-horizon reasoning through recursive self-improvement via dual-loop architecture and autonomous context updating, achieving significant performance gains on complex research benchmarks.