Anthropic ships gov AI; OpenAI agents advance; Regulators…

Anthropic / Claude ecosystem

Claude Fable 5 transitions to pay-per-use pricing

Anthropic's Claude Fable 5, previously included with subscriptions, will now move to a premium pay-per-use model starting July 12, 2026. Users have two days to utilize remaining subscription access.

Anthropic launches Claude for Government with Code & Cowork on FedRAMP High

Anthropic has released 'Claude for Government', deploying its Code and Cowork agentic AI products in public beta on FedRAMP High. This marks the first time finished agentic AI tools are available for direct procurement and use in US government knowledge work, complete with audit logging.

Claude Code desktop launches built-in sandboxed browser

Anthropic has integrated a built-in, sandboxed web browser directly into Claude Code on desktop. This new feature allows developers to access documentation, designs, and web applications without needing to switch between windows or open a separate browser like Chrome.

Developers face RCE via Claude Code 'auto-mode' exploit

Researchers from the AI Now Institute have demonstrated a remote code execution (RCE) exploit in Claude Code's auto-mode, affecting versions 2.1.116–2.1.199. Untrusted text in code repositories can reach AI coding agents, allowing RCE across Claude Sonnet 4.6, Sonnet 5, and Opus 4.8 without modification.

Frontier model providers

OpenAI launches GPT-5.6 and ChatGPT Work for agentic workflows

OpenAI has released GPT-5.6 and 'ChatGPT Work', transitioning from a chatbot to a unified desktop application for agentic workflows. This new system consolidates Chat, Work, and Codex functionalities, offering autonomous task execution and real-time application integration.

OpenAI's GPT-5.6 Sol Ultra claims proof of 50-year-old math conjecture

OpenAI's GPT-5.6 Sol Ultra claims to have generated a complete proof of the 50-year-old unsolved Cycle Double Cover Conjecture in graph theory in under one hour. The proof is currently pending peer review verification.

AI developer tooling & infrastructure

Cursor adds side chats and transcript search for smoother coding workflows

Cursor has introduced new 'Side Chats' and 'Transcript Search' features to its coding assistant. These additions aim to help developers manage complex coding workflows by providing parallel AI conversation threads and searchable chat history, preventing loss of context.

GitHub Copilot redesigns code review with Unix-style tools for cost reduction

GitHub has demonstrated that redesigning its Copilot code review feature with simpler, constrained Unix-style tools improves review quality and significantly reduces operational costs compared to more complex tooling.

MCP Servers 2026.7.10 released, ensuring stability for enterprise AI agents

The Model Context Protocol (MCP) Servers 2026.7.10 has been released, providing synchronized maintenance updates to core protocol servers including Filesystem, Time, Fetch, and Git. This ensures stability and compatibility across enterprise AI agent infrastructure.

CrewAI 1.15.2 strengthens multi-agent orchestration for QA engineers

CrewAI 1.15.2 has been released, bringing significant improvements to multi-agent orchestration. New features include inline skills, templated Flow inputs, repository agents, and enhanced streaming protocol for enterprise-scale AI workflow automation.

LangChain launches OpenWiki 0.1.0 for proactive AI agent memory

LangChain has introduced OpenWiki 0.1.0, a new proactive memory system designed to enhance the capabilities of AI agents. This system aims to provide agents with a more robust and accessible knowledge base.

Cloud & platform providers

Amazon EC2 G7 instances now available in AWS US East (N. Virginia)

Amazon Web Services (AWS) has announced the general availability of its new EC2 G7 instances in the US East (N. Virginia) Region. These instances offer up to 4.6 times faster AI inference performance and up to 2.1 times faster graphics performance compared to previous G6 instances.

Stardog and AWS partner to build semantic layer for agentic AI on Amazon Bedrock

Stardog and Amazon Web Services (AWS) have partnered to demonstrate the deployment of Stardog's semantic layer and knowledge graph federation with Amazon Bedrock AgentCore. This integration enables agentic AI systems to reason over diverse enterprise data without complex ETL processes.

AWS and Unsloth demonstrate quantized model deployment on Amazon SageMaker AI

Amazon Web Services (AWS) and Unsloth have collaborated to demonstrate four deployment patterns for quantized models on Amazon SageMaker AI. This initiative aims to reduce serving costs and facilitate flexible model compression workflows for AI applications.

Cloudflare turns AI bug hunting into a vulnerability pipeline

Cloudflare has developed 'Project Glasswing', an orchestrated multi-agent vulnerability pipeline and validation system. This research demonstrates that AI security work is enhanced by treating models as replaceable compute engines with persistent state management and independent validation, rather than relying solely on longer prompts.

AI policy, regulation & governance

Senator Markey releases 'The AI Accountability Agenda' targeting Big Tech

Senator Edward J. Markey has unveiled a comprehensive policy framework titled 'The AI Accountability Agenda: Taking Power Back from Big Tech'. This agenda outlines six key priorities, including worker protections, child safety, civil rights, healthcare, environmental impacts, and wealth distribution, aiming to address AI harms.

Senator Markey introduces federal AI accountability package, including data centre certification

Senator Ed Markey has released a federal AI accountability package, proposing bills on data centre certification, algorithm bias audits, employer AI hiring restrictions, and chatbot safeguards for children. This initiative aims to establish federal guardrails against a fragmented landscape of state-level regulations.

Australia's CISC proposes SOCI Act reforms to combat AI-driven cyber threats

The Australian Cyber and Infrastructure Security Centre (CISC) is advancing 21 proposed reforms to the Security of Critical Infrastructure Act 2018 (SOCI Act). These reforms aim to address emerging AI-enabled cyber threats and modernize the protection of critical infrastructure.

OpenAI suspends API access for Alibaba over suspected model distillation

OpenAI has suspended API access for the blacklisted Chinese firm Alibaba, citing suspected model distillation. This action exposes regulatory gaps in AI export controls and raises questions about compliance in global AI services.

Industry & market moves

Paris Voice AI Startup Gradium Hits $100M Seed Round with NVIDIA Backing

Paris-based voice AI startup Gradium has secured a $100 million seed round with backing from NVIDIA, marking one of Europe's largest seed rounds in AI history. The funding aims to help Gradium compete with established players like ElevenLabs and Google in the real-time voice AI market.

EQT to acquire Copia Power, an integrated power and AI infrastructure platform

EQT Infrastructure VII is set to acquire Copia Power, a leading integrated power and AI infrastructure platform. This acquisition expands EQT's portfolio of AI infrastructure investments, directly addressing the energy bottleneck in US data center growth.

Apple sues OpenAI for alleged trade secret theft via poached employees

Apple has filed a lawsuit against OpenAI, alleging that the ChatGPT maker systematically stole trade secrets about unreleased products. Apple claims OpenAI recruited over 400 employees and encouraged them to share confidential materials during their notice periods.

AI product & feature launches

Humanoid robots perform first-ever live surgery in UC San Diego trial

Humanoid robots have successfully performed live surgical procedures for the first time in a trial conducted by the UC San Diego Center for the Future of Surgery. The publication in Nature highlights a compact and deployable alternative to traditional large surgical robotic systems.

Biomedical AI agent Biomni speeds up scientific research, deployed in 10,000+ labs

Phylo, founded by Kexin Huang, has launched Biomni, the first general-purpose biomedical AI agent capable of autonomously executing complex research tasks in minutes that would typically take human scientists weeks. A prototype is already deployed across over 10,000 labs.

Research with immediate practical relevance

China's Orca world model matches specialized robotics systems without action labels

The Beijing Academy of Artificial Intelligence (BAAI) has released Orca, a world model (0.8B and 4B parameters) that demonstrates it can match specialized robotics systems across multiple tasks, even when trained without explicit action labels. This approach potentially mitigates robotics' chronic data scarcity issues.