Coding agent safety brakes; defense AI funding soars

Anthropic / Claude ecosystem

Anthropic Introduces India Pricing For Claude AI Plans

Anthropic launches localized INR pricing for Claude Pro and Team plans, removing foreign exchange charges and enabling predictable billing for Indian developers and enterprises.

Claude Code Auto Mode Now Catches the Empty-Variable rm -rf

A new guard in Claude Code's auto mode detects and pauses execution of destructive commands like 'rm -rf' when the target path is derived from an empty or unresolved variable, preventing accidental system wipes.

Claude Code users keep 50% higher limits until July 19

Anthropic extends a promotion providing a 50% increase in weekly usage limits for eligible Claude Code plans, now running until July 19, 2026.

Frontier model providers

No significant new developments.

AI developer tooling & infrastructure

Cross-Harness Tool Parity: One Config, Six AI Coding Environments

A developer introduces a unified configuration kernel that delivers byte-for-byte identical tooling across Claude Code, Cursor, Codex, Gemini CLI, Copilot, and Windsurf, enabling consistent AI-assisted coding workflows.

Agent Harness ECC Tops 228K Stars: Free Tool Turns Claude Code Into a Full Dev Team

The open‑source Agent Harness ECC provides 67 agents and 278 skills with a built‑in security scanner for Claude Code, turning it into a production‑ready development workflow.

Developers are shipping their own kill switch for coding agents that run 'rm -rf'

An open‑source Destructive Command Guard intercepts dangerous shell commands from AI coding agents, supporting multiple CLIs with sub‑millisecond performance.

Prime Intellect Releases Verifiers v1: Composable Tasksets, Harnesses, and Runtimes for Agentic RL Training and Evaluations

Prime Intellect launches Verifiers v1, a composable environment stack for agentic reinforcement learning, including tasksets, evaluation harnesses, and execution runtimes.

Cloud & platform providers

Qwen3 embedding and reranking models for retrieval are now available in Amazon SageMaker JumpStart

AWS adds Alibaba's Qwen3‑VL‑Embedding‑2B and Qwen3‑Reranker‑4B models to SageMaker JumpStart, enabling multimodal retrieval and cross‑modal reranking pipelines.

AI policy, regulation & governance

Malaysia advances with stakeholder engagement on its first AI governance bill

Malaysia actively engages stakeholders to shape its first cross‑sector AI law, aiming to establish guardrails before a global consensus emerges.

US Eases AI Chip Export Rules for UAE

The U.S. relaxes export controls, allowing UAE subsidiaries of Apple, Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI to receive advanced AI chips under a License Exception Strategic Trade Authorization.

Australian artists and activists protest a cabinet‑level plan linking a $50 billion AI data‑centre investment to a $350 million fund for artists, fearing it may erode copyright protections.

Industry & market moves

Valarian raises $50 million to help governments and enterprises escape America's cloud grip

Valarian secures $50 million in funding to advance technology sovereignty and alternative cloud strategies for European governments and enterprises, reducing reliance on US providers.

Helsing Raises $1.8 Billion, Reaches $18 Billion Valuation in Major Defense AI Milestone

European defense AI startup Helsing raises $1.8 billion in funding, achieving an $18 billion valuation, marking one of the largest private investments in defense technology.

AI product & feature launches

Minimax 2.7 Goes Open-Source, DeepSeek V4 on the Horizon

Chinese AI company Minimax releases version 2.7 of its model under an open-source license, and hints that DeepSeek V4 is forthcoming.

Notion's Ship OS Bets the Team Layer Is the Real AI Harness

Notion plans to release Ship OS, an agent‑native platform designed to run the entire product development cycle within Notion, leveraging AI agents and team collaboration.

Global consortium releases AI toolbox to unlock new Alzheimer's treatments

The C‑BRAIN consortium launches three open‑source AI tools—AI Literature and Data Synthesis, Dark Data Analyzer, and Reviewer Three—to help researchers analyze biomedical literature, surface hidden data, and provide peer‑review‑style feedback for neurodegeneration research.

Cybersecurity startup Reken emerges from stealth with on-device shield for AI-powered fraud

Reken comes out of stealth with Northstar, an on‑device AI security platform that defends against AI‑powered phishing, deepfakes, and scams without sending data to the cloud.

Research with immediate practical relevance

Brain Scan AI Beats GPT-5 by Over 20 Points in Real-Week Hospital Trial

A specialist medical AI model, NeuroVFM, pre‑trained on 5.24 million clinical MRI and CT volumes without hand‑labeled data, outperforms GPT‑5 by over 20 accuracy points in a one‑week hospital feasibility study.

MuScriptor: Open Audio-to-MIDI Model for Full Mixes

Kyutai releases MuScriptor, an open‑weight model that converts full audio mixes into per‑instrument MIDI tracks with state‑of‑the‑art note‑level accuracy.