Agents drive infra and policy; funding pours in

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

Introducing the Voice Agent Builder | xAI

xAI has launched its Voice Agent Builder, a no-code platform that allows users to create production-ready voice agents in under two minutes. The platform integrates telephony, knowledge retrieval, and observability features, aiming to simplify the deployment of complex voice AI solutions.

Mistral AI updates Leanstral for Lean 4 proof work | LavX News

Mistral AI has released Leanstral 1.5, a specialized 119B-parameter sparse model optimized for Lean 4 formal proof engineering, autoformalization, and theorem proving. This update offers a 256K-token context window, making it highly effective for complex mathematical and logical reasoning tasks.

Safely Releasing Frontier Models to Customers | Artificial Intelligence

AWS and Anthropic are collaborating to release Claude Fable 5 models with enhanced guardrails on Amazon Bedrock. This initiative focuses on balancing the power of frontier AI with robust security measures, aiming to prevent malicious actors from accessing advanced cybersecurity tools while empowering defenders.

AI developer tooling & infrastructure

Why we built ADK 2.0 - Google Developers Blog

Google has released ADK 2.0, introducing a structured workflow runtime that combines deterministic execution with LLM-driven agent capabilities. This update, now available in Go alongside existing Python support, aims to facilitate the development of production-grade AI applications by providing a robust framework for agent orchestration.

Google launches remote MCP server for Gemini Enterprise

Google has introduced a remote Model Context Protocol (MCP) server for its Gemini Enterprise Agent Platform. This enables external AI agents and development tools to securely integrate with Google Cloud resources, allowing third-party agents to access models, prompts, and notebooks while maintaining IT governance through Cloud IAM Deny policies.

Reduce Token Use and Response Time in Your LangChain Agent With Airbyte | Airbyte

Airbyte has launched its Context Store for LangChain agents, a managed, searchable data index layer that significantly reduces token usage by up to 80% and tool calls by 40%. This optimization improves the efficiency and cost-effectiveness of LangChain agent deployments.

MCP 2026-07-28: what changes for AI agent authentication

The upcoming MCP protocol update, effective July 28, 2026, removes session-based trust and adopts stateless OAuth 2.1 authorization for AI agent authentication. This requires enterprises to audit and migrate any session-dependent MCP deployments to avoid token replay and confused-deputy risks, fundamentally altering how agents secure their access.

OpenCV Debuts Cloud-Optimized Build and Live Announcement | Let's Data Science

OpenCV has debuted COOL (Cloud-Optimized OpenCV), a cloud-optimized native build designed to reduce CPU cost and increase throughput for large-scale computer vision inference on Arm-based cloud instances like AWS Graviton. This aims to make computer vision more efficient and accessible in cloud environments.

Cloud & platform providers

AWS Announces General Availability Of AI Agents, Investment In FDEs – SMBtech

AWS has announced the general availability of AI Agents for WorkSpaces, allowing AI agents to operate legacy desktop applications without requiring API migration. This addresses a significant last-mile automation challenge for enterprises looking to integrate AI into existing, non-API-enabled workflows.

AWS details how AG-UI brings generative agent interfaces to Amazon Bedrock AgentCore

AWS has standardized agent-to-frontend communication through the open AG-UI protocol, alongside the release of FAST v0.4.1 (Fullstack AgentCore Solution Template). This initiative decouples rich UI interactions from backend orchestration frameworks, aiming to reduce the brittleness of streaming parsers and streamline generative agent interface development.

Cloudflare's new policy pushes AI companies to pay for publishers' content | TechCrunch

Cloudflare is implementing a new policy, effective September 15, 2026, to force AI companies to separate search crawlers from training crawlers. This move effectively requires AI developers to pay or license content from publishers, using Cloudflare's default platform settings to ensure content creators are compensated for their work.

AI policy, regulation & governance

Fable and Mythos: Anthropic says US lifts export ban on its advanced AI tools

Anthropic announced that the US Department of Commerce has reversed its export ban on advanced Claude models, Fable 5 and Mythos 5. This decision follows Anthropic's agreement to implement enhanced security safeguards and collaborate with the US government on future AI releases, restoring global access to these models after weeks of restrictions.

Industry & market moves

The DeepMind Trio Who Built a Poker AI Are Now Making Money for Quant Hedge Funds

Former DeepMind poker AI researchers have founded EquiLibre Technologies, applying reinforcement learning to algorithmic trading. The company has achieved zero negative months since its inception and secured a $500M Series A funding round, demonstrating the successful transfer of advanced AI research to high-stakes financial applications.

Meta Is About to Make Its Next Billion-Dollar Bet. Wall Street Thinks It Could Be Huge  - 24/7 Wall St.

Meta is reportedly organizing an AI cloud computing rental business, leveraging its substantial $145 billion infrastructure investment. This strategic move could position Meta as a fourth hyperscale cloud provider, competing directly with established giants like AWS, Azure, and Google Cloud.

Announcing our $800M Series C to accelerate the shift to open-source AI

Together AI has raised $800M in Series C funding to significantly scale its open-source AI infrastructure and production inference capabilities. The funding includes commitments for 500 MW of compute capacity from investors like Aramco Ventures, NVIDIA, and Vista Equity, aiming to accelerate the shift towards open-source AI adoption.

Datadog Acquires Adaptive ML to Accelerate Its Investment in AI Research and Development |

Datadog has acquired Adaptive ML to accelerate its AI research and development, specifically focusing on world models and agentic LLM post-training for observability. This acquisition aims to enhance Datadog's capabilities in monitoring and understanding complex AI systems.

Thailand's Amity Robotics raises $7M to expand AI concierge robots across Asia - TNGlobal

Thailand-based Amity Robotics has closed a $7 million seed funding round to expand its AI concierge robot deployments across Asia. The company has already demonstrated commercial viability with successful deployments in over 30 properties, indicating a growing market for physical AI solutions in hospitality and services.

Genesys Expands Contact Center AI Ambitions With Acquisition of Pinkfish | TMC Insight

Genesys has acquired Pinkfish to integrate over 25,000 MCP server tools and 500 prebuilt enterprise integrations, accelerating agentic AI for contact center automation. This move signifies the market's shift towards autonomous workflows and Genesys's commitment to enhancing customer experience through AI.

GapMinder leads USD 3 million seed round in Slovenian AI software startup Codeplain  | Romania Insider

GapMinder has led a $3 million seed funding round into Codeplain, a Slovenian AI software startup. This investment expands GapMinder's portfolio of Eastern European technology and AI companies, signaling continued investor interest in regional AI innovation.

AI product & feature launches

Meta is adding ridiculous ‘rate limits’ and a soft paywall to its smart glasses | The Verge

Meta is introducing a new $19.99/month subscription tier, Meta One Premium, for expanded access to features on its smart glasses, such as Conversation Focus. This move implements rate limits and a soft paywall even for features that run locally on-device, raising questions about Meta's monetization strategy for its AI hardware.

Instagram Launches Spin View and Multi-cam Features for Ray-Ban Meta, Oakley Meta, Meta Glasses

Instagram has launched new interactive Spin View and Multi-cam features specifically for Meta's smart glasses (Ray-Ban Meta, Oakley Meta, and Meta Glasses). These tools enable creators to capture and share immersive, multi-angle stories with synchronized footage and new editing capabilities, enhancing hands-free content creation.

XBTFX Launches AI Trading Stack Connecting Compatible AI Agents to Live Crypto and CFD Trading Accounts | CoinCodex

XBTFX has launched its AI Trading Stack, a platform that bridges AI analysis with live trading execution by making trading accounts programmable through structured, authenticated access to account data and market infrastructure. The stack includes an MCP Server, Skills Hub, and Trading API.

Proton takes on ChatGPT and Claude with launch of Lumo 2.0 - UK Tech News

Proton has launched Lumo 2.0, a privacy-first AI assistant designed to compete with OpenAI's ChatGPT and Anthropic's Claude. Lumo 2.0 claims to offer comparable capabilities while ensuring all conversations are encrypted and never used for model training, prioritizing user data privacy.

Research with immediate practical relevance

Introducing GeneBench-Pro | OpenAI

OpenAI has introduced GeneBench-Pro, a new research-level benchmark designed to evaluate AI models' ability to handle complex, judgment-heavy analysis in computational biology. This benchmark addresses the current gap in assessing system-level scientific reasoning skills, pushing AI capabilities beyond mere code fluency for scientific applications.

Pulpie: Pareto-Optimal Models for Cleaning the Web

Feyn Inc. researchers have released Pulpie, a family of open-source HTML content extraction models that achieve state-of-the-art quality at 1/20th the inference cost. These models utilize an efficient encoder architecture, making web data cleaning and extraction significantly more economical.

A Self-Evolving Agent Framework That Treats Hardware Design as Repository-Level Code Evolution (Nvidia Research)

Nvidia Research has introduced HORIZON, a self-evolving agent framework that extends repository-scale self-evolution from EDA software to hardware design artifacts. This framework achieved 100% benchmark completion on ChipBench, RTLLM, and Verilog-Eval with a fully hands-free agentic loop, demonstrating autonomous hardware design capabilities.

‘BioShocking’ jailbreak tricks AI browsers into disclosing private data | news | SC Media

LayerX researchers have demonstrated a 'BioShocking' jailbreak attack that exploits AI browsers' contextual understanding to bypass safety guardrails and extract sensitive credentials. This vulnerability highlights new risks in AI agent security and the need for advanced protective measures.

Google AI Introduces TabFM: A Hybrid-Attention Tabular Foundation Model for Zero-Shot Classification and Regression - MarkTechPost

Google AI has introduced TabFM (Tabular Foundation Model), a zero-shot tabular foundation model capable of performing classification and regression on unseen tables without prior training, tuning, or feature engineering. This is achieved through in-context learning, streamlining data analysis workflows.