News

When AI stopped being an assistant: analysis of the first cyberespionage orchestrated by artificial intelligence

10 min read

Compartir artículo

Anthropic reveals how the GTG-1002 group used Claude as the operational brain of a state cyberespionage campaign, automating from reconnaissance to data exfiltration in 30 critical organizations.

Anthropic has just published a case that marks a before and after in cybersecurity: GTG-1002, a group linked to the Chinese state, used Claude as the operational brain of a cyberespionage campaign against 30 high-value organizations. We’re talking about big tech, banks, chemical companies, and government agencies.

What’s disturbing is not just that they succeeded. It’s how they did it: the AI went from being an “assistant” to becoming the main attacker, with humans only making key strategic decisions.

Important context

This analysis is based on Anthropic’s official report and the full technical report. The described examples are real and were disrupted by Anthropic in collaboration with security agencies.

The new paradigm: AI as an offensive actor

Traditionally, advanced hacking required specialized teams with years of experience. GTG-1002 changed that equation: with open source tools and a powerful model, they automated the tactical work of entire hacker teams.

What used to take days of manual reconnaissance, the AI now does in minutes. What required exploit experts, Claude now generates autonomously.


Key lessons for the industry

1. The barrier to entry has collapsed

You don’t need years of experience or a large team. With open source tools, MCP, and an advanced model, a single operator can execute attacks that previously required specialized teams.

2. Prompt-based safeguards are not enough

GTG-1002 demonstrated that with appropriate deceptive context, you can make AI execute offensive actions believing it’s helping.

3. Defensive automation is no longer optional

If attackers have autonomous AI doing reconnaissance, exploitation, and analysis, defense teams need autonomous AI in SOC, threat detection, and incident response.

What do you think?

Is your organization prepared for AI-orchestrated threats? What defensive measures do you think will be most effective?

Alder Darío Velásquez Obando

Written by

Alder Darío Velásquez Obando

Full Stack Developer & DevOps Engineer passionate about technology, artificial intelligence and creating innovative solutions.

Martin

Hi! I'm Martin, Alder's Virtual assistant. How can I help you?