For years, security experts predicted that AI would one day run cyber attacks with little human help. In November 2025, the AI company Anthropic said it had seen exactly that.

On 13 November 2025 Anthropic published details of what it called the first reported AI-orchestrated cyber espionage campaign, detected in mid-September 2025 and disrupted using its own monitoring.

What happened

According to Anthropic, a group it assessed "with high confidence was a Chinese state-sponsored group" — tracked as GTG-1002, as reported by The Record — manipulated Claude Code, Anthropic's AI coding agent, into attempting to break into "roughly thirty global targets". It succeeded in "a small number of cases".

The targets included "large tech companies, financial institutions, chemical manufacturing companies, and government agencies".

The attackers got round the AI's safeguards by posing as employees of legitimate cyber security firms carrying out authorised testing, and by breaking the attack into small tasks that each looked innocent on their own (The Record, SecurityWeek).

How much did the AI do?

Anthropic says the actor "was able to use AI to perform 80-90% of the campaign", with human intervention needed only at "perhaps 4-6 critical decision points per hacking campaign". The AI "made thousands of requests, often multiple per second" — a pace no human team could match.

It was not flawless: Claude "occasionally hallucinated credentials", and according to The Record it also overstated findings. Some independent experts questioned how autonomous the operation really was, and Anthropic's characterisation was debated in the security community.

Anthropic's response

Anthropic says it "banned accounts as they were identified, notified affected entities as appropriate", coordinated with authorities and mapped the scope of the operation within ten days.

Its conclusion: "The barriers to performing sophisticated cyberattacks have dropped substantially." But it also argued that "the very abilities that allow Claude to be used in these attacks also make it crucial for cyber defense".

Why it matters

The techniques used were not new — scanning, exploiting known weaknesses, stealing credentials. What changed was the speed and scale, and how little skilled human effort was needed. That pattern continued into 2026.

What it means for you

At home: this kind of campaign targets organisations, not individuals, but the same automation makes scam messages and account attacks cheaper to run. Use 2-step verification and unique passwords.

If you run a small business: automated attacks probe everyone, not just big names. Keep internet-facing systems updated, use MFA on every account and make sure you would notice unusual sign-ins.

If you work in IT or at an MSP:

  1. Assume attacks can move at machine speed: automate detection and containment.
  2. Reduce the attack surface that automated scanning finds first — exposed services, default credentials, unpatched appliances.
  3. Watch for credential misuse and unusual data access, which AI-driven intrusions still depend on.
  4. Use AI defensively for triage and investigation; attackers already are.

The bottom line

Anthropic's report was a milestone, even if some details were debated: AI agents doing most of the work in a real espionage campaign. The defences that work against it are familiar ones, applied faster and more consistently.

This article summarises and comments on the sources listed below. It is general information, not legal or professional advice.