When AI Becomes the Weapon

It seems like this is becoming too frequent.

A lone hacker used Claude and ChatGPT to breach Mexico’s government.

1 person.

Two AI subscriptions.

Nine agencies compromised.

150 gigabytes of data stolen.

The breached agencies include Mexico’s federal tax authority and the national electoral institute, also state level systems in the Mexican states of Jalisco, Michaocan and Tamaulipas.

The data stolen includes 195 million taxpayer records, voter registration files, government employee credentials and civil registry data.

This one person managed to disable Claude’s guardrails. And ChatGPT’s. This is what the Pentagon wants Claude to officially do, so that they can do these types without having to hack the Claude built in safety controls.

The AIs produced ready-to-execute plans, exact targets and credentials.

It produced operational attack playbooks, on demand.

When Claude figured out what the guy was doing, he switched to ChatGPT.

If this is one ONE PERSON can do, imagine what an entire government sponsored team can do using these same AI tools.

I have seen reports of hackers, using AI, moving laterally inside the network within 9 minutes of initial access.

What this means that the tools and techniques you were using to get this far WILL NOT WORK going forward.

If you don’t have a plan, please contact us; we are happy to assist.

Credit: Breached.Com

by