White House Declares AI Agents Offensive Cyber Weapons After Chinese Hackers Weaponize Claude Code
Plus: NVIDIA launches NemoClaw at GTC and AI trading bots are losing real money.
Good morning. In today’s edition:
The White House formally classifies AI agents as offensive cyber weapons — months after a Chinese state group used Claude Code to automate 80-90% of a cyberattack campaign
NVIDIA unveils NemoClaw at GTC 2026, positioning OpenClaw as “the operating system for personal AI”
Gartner estimates only 130 of thousands of AI agent vendors are building genuinely agentic systems
An OpenClaw trading bot sends $441,000 to a random address after a decimal parsing error
Here’s what you need to know
The White House’s Cyber Strategy for America, released March 6, formally designates AI agents as instruments of offensive cyber operations — the first time U.S. policy has done so. The strategy pledges to “rapidly adopt and promote agentic AI” for both defensive and offensive “disruption” operations.
The timing is not abstract. According to a Just Security analysis published March 16, Anthropic assessed last November that a Chinese state-sponsored group had jailbroken Claude Code to automate 80-90% of a major cyber campaign targeting roughly 30 organizations across multiple countries. The agent handled reconnaissance, vulnerability identification, exploit crafting, and multi-target coordination. Human operators managed roughly 10-20% of the attack chain — primarily target selection and final authorization.
Separately, CVE-2026-21852 revealed that malicious repos could steal Anthropic API keys from Claude Code before users confirmed a repository as trusted. The vulnerability exploited the timing gap between when a developer opens a repo and when they explicitly trust it. Both stories land in the same week that OpenAI and Anthropic launched competing AI security agents within 14 days of each other — Codex Security on March 6, Claude Code Security on February 21.
What else we’re tracking
Cisco’s State of AI Security 2026 report puts the enterprise readiness gap in hard numbers: 83% of businesses are deploying agentic AI, but only 29% feel prepared to secure those deployments.
Abacus AI shipped “Secure OpenClaw” — SOC 2-certified sandboxed execution with persistent memory — a full month before NVIDIA announced NemoClaw. The enterprise OpenClaw race started earlier than the GTC headlines suggest.
Meta is preparing its largest layoffs since 2022, cutting more than 20% of its workforce to fund AI infrastructure. The stock rose 3%.
Top Stories
1. NVIDIA launched NemoClaw at GTC 2026, an open-source enterprise stack wrapping OpenClaw in sandboxing, privacy routing, and policy-based security guardrails. Jensen Huang called OpenClaw “the operating system for personal AI” in front of 25,000 developers at the SAP Center in San Jose. Partnerships with CrowdStrike and Cisco anchor the security layer. Full story.
2. Gartner estimates only ~130 of thousands of AI agent vendors are building genuinely agentic systems. The rest are chatbots and workflow automations with “agent” on the landing page. The agentic AI market is projected to hit $52 billion by 2030 — and Gartner separately predicts over 40% of agentic AI projects will be canceled by end of 2027. Full story.
3. AI trading agents deployed via OpenClaw produced their first documented financial disasters. A token distribution bot sent $441,000 to a random address after a decimal parsing error. A controlled experiment showed GPT-5 losing 62% of its trading capital in 17 days. Full story.
4. Anthropic was named TIME’s “Most Disruptive Company” after refusing a Pentagon military AI contract over ethics clauses. Claude Code is tracking toward $2.5 billion in annual revenue; Anthropic’s valuation has surpassed Goldman Sachs at $380 billion. Full story.
Also worth reading...
— The New Claw Times