Models and agents ship; EU labeling rules near

Anthropic / Claude ecosystem

Anthropic Research Introduces GRAM For Isolating Dangerous AI Knowledge

Anthropic and AE Studio present GRAM, a method using gradient-routed auxiliary modules to isolate and remove dangerous knowledge from AI models without degrading overall performance.

Frontier model providers

OpenAI rolls out GPT-5.6 and announces ChatGPT Work

OpenAI launches GPT-5.6 after government clearance, introducing ultra mode for multi-agent tasks, and unveils ChatGPT Work, an across-app assistant for generating documents, spreadsheets, and apps.

Introducing Muse Spark 1.1

Meta releases Muse Spark 1.1, a major upgrade with improved multimodal reasoning, tool use, coding, and a paid API, positioning it as a competitor to Anthropic and OpenAI.

China's DeepSeek developing its own AI chip, sources say

DeepSeek is reportedly building an in-house AI inference chip to reduce reliance on Nvidia and Huawei, backed by a major funding round.

AI developer tooling & infrastructure

Kastor brings infrastructure-as-code habits to AI agents

Kastor introduces a tool to define AI agents with typed specs and git-reviewable contracts, generating runnable projects akin to Terraform for agents.

Cloud & platform providers

We're rolling out AlphaEvolve widely to solve Google Cloud customers' hardest problems

Google Cloud makes AlphaEvolve, an automated algorithm optimization tool that outputs human-readable code, generally available to all customers.

Enhancing enterprise inference on Amazon SageMaker HyperPod with data capture, Hugging Face, NVMe, and Route 53 integration

AWS adds multi-tier data capture and direct Hugging Face model deployment to SageMaker HyperPod inference, with improved performance via NVMe and Route 53.

Microsoft's Secret Weapon for Cloud Security: AI Agents That Work in Hours, Not Weeks

Microsoft deploys a multi-agent AI system internally to review and harden cloud services, reducing security validation from weeks to hours.

Microsoft Pilots Agent ROI Dashboard in Foundry, Forms Copilot Hits Global Availability

Microsoft introduces a private ROI-tracking dashboard in Foundry and makes Forms Copilot generally available.

AI policy, regulation & governance

AI Content Labeling Enforcement Begins in 24 Days as EU Clears Compliance Code

EU enforcement of AI content labeling under the AI Act begins Aug 2, 2026, with multi-layer labeling requirements for AI-generated content.

Industry & market moves

HUMAIN and Cohere Launch Strategic Sovereign AI Compute Collaboration

HUMAIN and Cohere form a multi-year collaboration to scale frontier model development and enterprise applications, including Arabic-language and domain-adapted models.

AI unicorn Mercor acquires Deeptune after founder Brendan Foody backed the startup

Mercor acquires Deeptune to enhance its AI training simulators, consolidating its position in the AI training infrastructure market.

Mukesh Bansal-led Nurix AI acquires Verloop to expand enterprise offerings

Nurix AI acquires Verloop, combining voice and chat automation to expand enterprise customer engagement solutions.

Reevo Strengthens Its AI-Powered Prospecting and Outreach With Acquisition of Ciro

Reevo acquires Ciro to integrate its AI prospecting agent into Reevo's Revenue Operating System.

AI product & feature launches

Robbyant Unveils LingBot-World 2.0: Pioneering Hour-Long Real-Time Generation in World Models

Robbyant releases LingBot-World 2.0, an open-source world model enabling hour-long real-time generation with high-definition output.

Burn2kicad: AI Agents Cross the Chasm from Code to Circuit Board Design

A new tool allows AI agents to translate natural language prompts into manufacturable KiCad circuit board designs.

Research with immediate practical relevance

Open-Source World Model AlayaWorld Sustains Interactive Play Past the One-Minute Mark

AlayaWorld, an open-source framework, delivers 60+ seconds of coherent, interactive environments for embodied AI research.

OpenAI's AI beats every human at AtCoder, a top competitive programming contest

OpenAI's system solved all five problems at the AtCoder World Tour Finals 2026, dominating the competition.

Co-LMLM lets LMs query knowledge continuously

A new method keeps facts outside model weights and enables continuous, text-backed retrieval during generation, improving factual attribution.