Frontier models go autonomous; Anthropic opens portal

Anthropic / Claude ecosystem

Anthropic Opens Hard Questions Portal for Public Transparency

Anthropic invites public questions about its AI systems and commits to publishing reasoned answers, including uncertainty.

Claude Code Now Has Built-in Browser on Desktop

Claude Code's latest update introduces a sandboxed in-app browser, allowing developers to browse and interact with web pages without leaving the editor.

Developers Face RCE via Claude Code Auto-Mode Exploit

A host-level remote code execution vulnerability has been disclosed in AI coding assistants, enabling attacker-controlled code execution from injected prompts during automated security reviews.

Frontier model providers

GPT-5.6 Sol Autonomously Post-Trains Smaller Luna Model

OpenAI's GPT-5.6 Sol model autonomously optimized and post-trained the smaller Luna model from an underspecified prompt, significantly reducing manual training effort.

OpenAI's GPT-5.6 Sol Ultra Produces Proof of 50-Year-Old Math Conjecture

GPT-5.6 Sol Ultra reportedly generated a proof for the Cycle Double Cover Conjecture in under an hour using 64 subagents.

Meta Pulls Muse Image AI After Hollywood Privacy Backlash

Meta abruptly removed its Muse Image AI tool from Instagram following privacy concerns and negative feedback from Hollywood.

AI developer tooling & infrastructure

Cursor Prepares General-Purpose AI Agent Sand to Rival Claude and ChatGPT

Anysphere's Cursor is developing Sand, a workplace AI agent aiming to compete with Anthropic's Claude Cowork and OpenAI's ChatGPT Work.

GitHub Copilot Code Review Costs Cut Using Unix Tools

GitHub found that using simpler Unix-style exploration tools around pull request evidence reduced costs without degrading review quality.

Cursor Adds Side Chats and Transcript Search for Enhanced Productivity

The AI-powered code editor Cursor introduces parallel AI conversations and searchable chat histories to improve context management.

MCP Servers 2026.7.10 Released with Specification Updates

A maintenance release of core MCP servers (Filesystem, Time, Fetch, Git) brings them in sync with the latest specification and runtime improvements.

CrewAI 1.15.2 Released with Dynamic LLM Retrieval and Inline Skill Definitions

The new version of CrewAI introduces dynamic model retrieval, inline skill definitions, and enhanced flow tooling for AI agent development.

Cloud & platform providers

Amazon EC2 G7 Instances Now Available in AWS US East (N. Virginia) Region

AWS brings G7 instances with NVIDIA RTX PRO 4500 Blackwell GPUs to its largest region, offering higher AI inference performance.

Microsoft Foundry Hosted Agents Hit General Availability

Azure AI Foundry's framework-agnostic hosted agents become generally available, integrating with Teams and Copilot.

AI policy, regulation & governance

Trump Administration Targets State AI Laws over Ideology

The FTC proposes a policy to regulate AI outputs altered for ideological objectives, seeking public input.

Australia Warns of Unexpected AI Behaviour During Safety Testing

Australian authorities cite unexpected AI behaviour in safety tests, prompting regulatory oversight without new legislation.

Australia's AI Safety Institute: Progress and Future Needs

Australia establishes a national AI Safety Institute under the National AI Plan to test and guide AI safety strategy.

AI Loophole: Google and OpenAI Supplied Services to Blacklisted Chinese Firms via Singapore

Reports reveal that Google and OpenAI provided AI services to Singapore-based subsidiaries of sanctioned Chinese tech companies.

Industry & market moves

Paris Voice AI Startup Gradium Raises $100 Million Seed Round with NVIDIA Backing

Gradium secures a massive $100 million seed round, one of Europe's largest, backed by NVIDIA.

Oxylabs Raises $130 Million to Expand AI Data Infrastructure

Lithuania-based web data infrastructure company Oxylabs raises $130 million at a $3.6 billion valuation.

EQT to Acquire Integrated Power and AI Infrastructure Platform Copia Power

EQT Infrastructure VII agrees to acquire Copia Power, expanding its portfolio into combined energy and AI infrastructure campuses.

Apple Sues OpenAI for Alleged Trade Secret Theft Through Employee Poaching

Apple files a lawsuit accusing OpenAI of a coordinated campaign to steal confidential technology by hiring former Apple employees.

AI product & feature launches

IIT Delhi Opens Enrolments for Agentic AI Certificate Programme

IIT Delhi launches a six-month online Advanced Certificate in Agentic AI to upskill engineers and product leaders.

Biomni AI Agent Speeds Up Biomedical Research Across 25 Domains

Biomni, a general-purpose biomedical AI agent, autonomously executes diverse research tasks across thousands of publications.

Research with immediate practical relevance

A symlink-based vulnerability in AI coding assistants could allow malicious code execution via deceptive project settings.

Humanoid Robots Perform First-Ever Surgery in UC San Diego Trial

Teleoperated humanoid robots successfully performed complete surgical procedures in preclinical trials.