Google counters DeepSeek; security clampdown on agents

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

Google to Release Gemini 3.5 Pro on July 17, Competing with DeepSeek V4

Google is set to launch Gemini 3.5 Pro, an upgraded model with improved math reasoning and image generation capabilities. The release is positioned as a direct challenge to DeepSeek V4, signaling intensifying competition in frontier AI models.

AI developer tooling & infrastructure

LlamaIndex CEO Unveils Document Context Layer for AI Agents

LlamaIndex has introduced a Document Context Layer designed to enhance AI agents' access to context from documents, improving retrieval-augmented generation and agentic workflows. The layer aims to simplify data integration for agentic applications.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

Singapore's MAS Publishes SAFR Framework to Govern Autonomous AI Agents in Finance

The Monetary Authority of Singapore, with industry partners, released a white paper outlining Safeguards for Agentic Finance at Runtime (SAFR), proposing real-time governance checkpoints for autonomous AI agents in financial services.

Vietnam Releases List of 46 High-Risk AI Systems, Establishes Compliance Roadmap

Vietnam's government issued Decision No. 33/2026, officially categorizing 46 AI systems as high-risk and setting compliance deadlines. The list marks the country's first formal step toward a risk-based AI regulatory framework.

Australian Attorney-General's Department Adopts Microsoft Copilot Chat and Google NotebookLM for Protected Data

Australia's AGD has certified and deployed Microsoft Copilot Chat and Google NotebookLM within controlled cloud environments, enabling use with protected-level government data under strict policy guardrails.

Industry & market moves

Alibaba Bans Claude Code Over Security Risks, Deploys Homegrown Qoder AI Tool

Alibaba has prohibited its employees from using Anthropic's Claude Code, citing security concerns, and is steering them toward its internal Qoder assistant. The move follows claims of hidden China-detection logic and reflects rising geopolitical tensions in AI tooling.

Inturai Ventures Signs LOI to Acquire AI Drone Command Platform DomeCommand

Canadian firm Inturai Ventures plans to acquire DomeCommand, an AI-powered drone swarm command and control platform, to build a dual-use autonomy platform for defense and commercial applications. The deal includes a perpetual license for the Distri software platform.

Stakk Acquires ParaScript in $63M Deal to Form AI-Native Digital Trust Powerhouse

Australian-listed Stakk Limited completed the acquisition of ParaScript, combining its Digital Persona Graph with ParaScript's document intelligence to create one of the world's largest AI-native Digital Trust platforms for regulated industries.

Microsoft and Inception42 Bring Sovereign AI Agents to Microsoft 365 Copilot in UAE

In partnership with Inception42, Microsoft is enabling sovereign AI agents within Microsoft 365 Copilot for UAE enterprises, ensuring data remains in-country while leveraging AI assistance. The deployment addresses data sovereignty concerns for government and regulated industries.

AI product & feature launches

Synthetic Sciences Releases OpenScience, an Open-Source Alternative to Claude Science

OpenScience is a model-agnostic, open-source AI workbench for machine learning, biology, physics, and chemistry research, positioned as an open alternative to Anthropic's Claude Science. It supports multi-domain scientific workflows with an extensible architecture.

Safari Introduces AI Debugging Server via MCP for Web Development

Apple's Safari browser now supports a Model Context Protocol server, allowing developers to connect AI assistants directly to the browser for performance debugging and error detection. The integration streamlines AI-assisted web development workflows.

Research with immediate practical relevance

Agent Skill Malware Targets Claude Code and OpenAI Codex With Scanner-Evasion Techniques

A new study reveals that malicious code embedded in AI agent skills can evade static scanners, posing a threat to AI coding assistants like Claude Code and OpenAI Codex. Existing detection tools fail to identify these disguised attacks, highlighting a gap in application security for AI agents.

Security Scan Reveals 332 Critical Vulnerabilities Across LlamaIndex, AutoGen, and CrewAI

A security analysis of three popular AI agent frameworks uncovered hundreds of critical vulnerabilities, including injection flaws and privilege escalation risks. The findings highlight significant security gaps in current agentic AI tooling that could be exploited by attackers.