AI Weaponized at Scale: FortiGate Mass Compromise, npm Supply Chain Worm Hits AI Tooling, Anthropic Caught Between Pentagon and Beijing

A Russian-speaking threat actor with "limited technical capabilities" used commercial generative AI to compromise 600+ FortiGate devices across 55 countries — no zero-days needed, just AI-assisted credential attacks on exposed management ports. Simultaneously, a supply chain worm dubbed SANDWORM_MODE is spreading through 19 malicious npm packages that harvest crypto keys, CI secrets, and API tokens, with a new twist: MCP server injection with embedded prompt injection specifically targeting AI coding assistants. On the geopolitical front, Anthropic is getting squeezed from both sides — the Defense Secretary has summoned CEO Dario Amodei over military use of Claude, while Anthropic is publicly accusing Chinese AI labs of systematically mining Claude's outputs.

Impact Score

9/10

Key Stories

CYBER

AI-Assisted Threat Actor Compromises 600+ FortiGate Devices in 55 Countries

The Hacker News / Amazon Threat Intelligence — First confirmed large-scale campaign where AI turned an unsophisticated actor into a capable mass-compromise operator. No exploits — just AI-assisted credential attacks at scale. Jan 11 – Feb 18, 2026.

CYBER

SANDWORM_MODE: npm Supply Chain Worm Targeting Developer Environments and AI Coding Assistants

The Hacker News / Socket — 19 malicious npm packages with worm propagation via stolen npm/GitHub identities. New vector: MCP server injection with prompt injection payloads targeting AI coding tools like Claude Code and Copilot. Direct relevance to Dong's dev environment.

POLICY

Defense Secretary Summons Anthropic's Amodei Over Military Use of Claude

TechCrunch — Pentagon pressure on Anthropic signals accelerating military AI procurement push. Amodei has previously been cautious about defense contracts — this summons suggests the US government is done waiting.

POLICY

Anthropic Accuses Chinese AI Labs of Mining Claude as US Debates AI Chip Exports

TechCrunch — Systematic model extraction by Chinese labs — Anthropic going public with this while chip export controls are being debated is a calculated political move. Shapes the narrative around export restrictions.

CYBER

APT28 Operation MacroMaze: Russia Targets Western Europe with Webhook-Based Macro Malware

The Hacker News / S2 Grupo LAB52 — Active Sep 2025 – Jan 2026. Classic spear-phishing with a modern twist — INCLUDEPICTURE fields pointing to webhook.site for beaconing. Low-tech, effective, hard to detect at perimeter. Europe-focused.

Analyst Take

The FortiGate story is the one to watch. Amazon Threat Intelligence documented a financially motivated actor with *limited technical skills* running a 600-device, 55-country compromise campaign — because AI handled the parts they couldn't. This is the threat model shifting in real time: the barrier to entry for mass compromise just dropped by an order of magnitude. You don't need to be APT28 anymore. You need a ChatGPT subscription and exposed management ports. Every organization with internet-facing admin interfaces is now a legitimate target for actors who previously couldn't mount this kind of operation. The SANDWORM_MODE supply chain worm is the second alarm — the fact that it's now injecting into MCP servers and prompting AI coding assistants means the attack surface has expanded to include the tools developers trust most. If Dong or anyone on the team uses AI coding assistants connected to live credentials or CI environments, that pipeline needs an audit. The Anthropic squeeze from both Pentagon and Beijing is a longer-term signal: the US government is moving toward mandatory defense cooperation for frontier AI labs, while China is clearly treating Claude outputs as training data. Anthropic's public accusation is strategic positioning ahead of chip export legislation. Watch for either a formal defense contract announcement or a very public refusal from Amodei in the next 30 days — either outcome reshapes the competitive landscape.