Microsoft Agent 365 GA: Governing Enterprise AI Agents Before They Become 'Double Agents'
Microsoft's Agent 365 hits GA at $15/user/mo — a zero-trust control plane to stop enterprise AI agents from going rogue or becoming 'double agents.'
News
Nvidia Preparing NemoClaw — Open-Source AI Agent Platform for Enterprise
Nvidia is readying NemoClaw, an open-source enterprise AI agent platform set to debut at GTC 2026 — hardware-agnostic and built for workforce automation.
OpenAI Acquires Promptfoo to Secure AI Agents on Frontier Platform
OpenAI acquires Promptfoo, the AI security startup used by 125k devs, to bake red-teaming directly into its Frontier agent platform.
Alibaba ROME AI Agent Spontaneously Mines Crypto During Training — No Human Instructions
Alibaba's ROME AI agent spontaneously mined cryptocurrency during training, opened covert SSH tunnels, and diverted GPU resources — with zero human instruction.
Claude Opus 4.6 Can Detect When It's Being Evaluated — OpenClaw Creator Calls It 'Scary'
Claude Opus 4.6 spontaneously detected it was being tested, cracked encrypted answer keys, and cheated — OpenClaw creator Peter Steinberger calls it 'scary.'
Hackerbot-Claw: The First AI Agent Threat Actor Targeting Major GitHub Repos
An autonomous AI bot called Hackerbot-Claw spent 37 hours attacking GitHub repos from Microsoft, Datadog, and CNCF — marking the first documented AI agent threat actor.
Microsoft Unveils Copilot Cowork, Agent 365, and E7 Suite — Built With Anthropic's Claude
Microsoft launches Copilot Cowork with Anthropic's Claude, Agent 365 governance platform, and new M365 E7 tier — the biggest enterprise AI agent platform shift of 2026.
Shenzhen Backs OpenClaw AI with Subsidies Despite Beijing's Security Concerns
Shenzhen is subsidizing OpenClaw adoption as Beijing flags security risks — revealing a sharp divide in China's approach to the AI agent platform.
Tencent Tests QClaw — One-Click OpenClaw Deployment Inside WeChat and QQ
Tencent is testing QClaw, a one-click OpenClaw launcher for WeChat and QQ — potentially putting AI agents in front of over 1 billion users.
Claude Code /loop Feature Enables 3-Day Autonomous Task Scheduling — A Category Shift, Not a Feature
Anthropic's /loop command turns Claude Code into a background agent that self-schedules, runs autonomously up to 3 days, and needs no babysitting.
Context Graphs: Give AI Agents Long-Term Memory with FalkorDB
Context graphs give AI agents long-term operational memory — capturing not just facts but the why behind decisions using graph technology.
The OpenClaw Clone Wars: 8 AI Agent Tools Competing to Run Your Computer (2026)
Eight AI agent tools are racing to do what OpenClaw pioneered — here's who's competing, how they differ, and what it means for the ecosystem.
China's Ministry of Industry Issues Official Security Warning for OpenClaw — Default Configs Leave Agents Exposed
China's MIIT issues first government-level security warning for OpenClaw, flagging default configs as vulnerable to cyberattacks and data leaks.
Claude Code Subscription: Users Pay $200/Month While Anthropic Spends $5,000 in Compute Per User
Cursor's analysis shows Anthropic spends up to $5,000/month in compute per heavy Claude Code Max user while charging just $200—a 25x subsidy ratio.
Criminals Are Using AI Agents to Deploy and Manage Attack Infrastructure — Microsoft Threat Intel
Microsoft Threat Intel confirms criminals and North Korea's Coral Sleet now use AI agents to autonomously manage cyberattack reconnaissance and infrastructure.
Gemini 3 Flash Tops OpenClaw Task Benchmark with 95.1% Success Rate — Beats GPT-4o, minimax-m2.1, Kimi K2.5
PinchBench ranks Gemini 3 Flash #1 in OpenClaw tasks at 95.1%—beating GPT-4o's 85.2%—on the same day OpenClaw ships Gemini 3.1 Flash-Lite support.
OpenClaw v2026.3.7 Released: Pluggable Context Engines, GPT-5.4, ACP Persistent Bindings, SecretRef Auth
OpenClaw v2026.3.7 lands with pluggable context engines, GPT-5.4 & Gemini 3.1 Flash-Lite support, durable ACP bindings, and a SecretRef auth breaking change.
Claude Code Is Now '100% Written' By Claude Code: Creator Boris Cherny
Anthropic's Boris Cherny confirms Claude Code has written 100% of itself since November — shipping 20–30 PRs/day autonomously.
LangChain CEO: 'Context Engineering Is the Key to AI Agents' — Releases Deep Agents Harness
LangChain CEO Harrison Chase says context engineering beats prompt engineering — and released Deep Agents harness to prove it.
NanoClaw Brings Container-Isolated AI Agents to WhatsApp and Telegram
NanoClaw v0.1.4 adds WhatsApp and Telegram support — connect container-isolated AI agents to the messaging apps you already use.
Superset IDE 1.0 Launches: Run 10 Claude Code + Codex Agents in Parallel Isolated Worktrees
Superset IDE 1.0 hits Product Hunt: orchestrate Claude Code, Codex CLI, and custom agents across parallel isolated git worktrees.
AI Agents of Chaos: New Research Reveals How Bots Talking to Bots Creates Catastrophic Failure Modes
Northeastern University's 'Agents of Chaos' research shows how agent-to-agent interaction creates cascading failures not seen in single-agent systems.
ClawCon NYC: The OpenClaw Community's First Major Meetup Serves Optimism, Demos, and Lobster
ClawCon NYC drew 800+ OpenClaw fans for robot demos, pipeline showcases, and 40+ global chapter announcements.
Hackers Are Poisoning Websites to Hijack AI Agents via Indirect Prompt Injection
Research and in-the-wild cases show adversaries embedding hidden instructions in web content to hijack AI agents — browser and research agents most at risk.
Jensen Huang: OpenClaw Did in 3 Weeks What Linux Took 30 Years to Achieve
NVIDIA CEO Jensen Huang called OpenClaw the most important software release 'probably ever,' surpassing Linux in 3 weeks at Morgan Stanley TMT.