AI agents boom; energy and security worries mount

Anthropic / Claude ecosystem

Claude Code adds cross-session messaging on macOS

Claude Code v2.1.224 introduces a feature allowing direct messaging between concurrent coding sessions on macOS and Linux, enabling context sharing without manual copying.

Claude service outage disrupts users

Anthropic's Claude AI experienced a service outage, impacting user access and workflows.

Frontier model providers

OpenAI plans hockey puck-sized AI speaker for 2027

OpenAI is developing a screenless, donut-shaped AI speaker with moving parts, expected to launch in 2027 at over $300, marking its first hardware device.

xAI launches Imagine Image 2.0 with precise editing, ranks second in Arena

xAI releases Grok Imagine Image 2.0, featuring region editing, multi-reference inputs, and templates, achieving second place behind OpenAI's GPT-Image-2 in Arena benchmarks.

AI developer tooling & infrastructure

GitHub Copilot introduces model-aware completions and new CLI commands

The latest Copilot update brings Auto detection details inside completions, /worktree and /rewind commands, and enhanced VS Code integration for AI-powered coding.

Cloud & platform providers

Cloudflare launches Kitesurf, a browser built for AI agents

Cloudflare unveils Kitesurf, a cloud-hosted headless browser optimized for AI agent tasks, claiming significantly less CPU and memory usage than Chromium, available in beta via Browser Run.

Azure API Management adds dedicated AI gateway tier in preview

Microsoft introduces a new AI Gateway tier in Azure API Management, enabling model-centric governance and multi-provider model hosting with MCP support.

Cloudflare releases Radar Researcher AI tool for Internet data exploration

Cloudflare launches Radar Researcher, an AI-powered tool that allows users to query Internet traffic and security data using natural language.

AI policy, regulation & governance

India examines Meta's recommendation system status under IT intermediary rules

Indian authorities are reviewing whether Meta's content recommendation algorithms qualify as an intermediary, which could affect its liability protections.

Australian proposal calls for public AI near-miss register

A proposal suggests establishing a national AI Safety Institute and a public 'near-miss' register for AI incidents to improve accountability and prevent harm.

Industry & market moves

xAI co-founder Yang Ge departs citing overwork

xAI co-founder Yang Ge has left the company due to excessive workload, adding to a series of high-profile departures from Elon Musk's AI venture.

Cloudflare raises full-year forecast on AI-driven spending

Cloudflare increased its annual revenue forecast, citing strong demand for its AI-related platform and security products, sending shares higher.

Nvidia to invest up to $3 billion in Stargate power developer Lancium

Nvidia is planning a strategic investment of up to $3 billion in Lancium, a developer of power and land infrastructure for Stargate AI data centres, signaling a move into the energy supply chain for AI compute.

AI product & feature launches

ScribeMe publicly integrates with Meta AI glasses for blind and low-vision users

ScribeMe launches public access to its hands-free AI description service through Meta AI glasses, enhancing accessibility for visually impaired users.

BASF deploys AI materials discovery platform from Orbital Industries

BASF's Environmental Catalyst division adopted Orbital Industries' CurieOS AI platform to accelerate automotive emission catalyst innovation.

Research with immediate practical relevance

Meta AI claims gold medals in reasoning Olympiads with no tools

Meta's latest AI model reportedly achieved five gold medals in pure reasoning STEM Olympiad competitions, including two perfect scores in physics theory.

DEF CON 34 exposes fundamental flaws in AI agent security architecture

Presentations at DEF CON 34 revealed systemic, unpatchable vulnerabilities in AI agent ecosystems, challenging the prevailing optimistic security narratives.

Khora scales multiplayer world models to 1,000+ players

Khora demonstrates a multi-agent world model with linear-time complexity, supporting over 1,000 concurrent players in a live public demo.

AI agents consume 600x more energy than simple chat prompts, study finds

A new analysis reveals that AI agents can use up to 600 times more energy per task than a basic chat query, raising sustainability concerns.