AI used in offensive operations targeting the Mexican government

Cyber News
,
1

Report describing how Claude and OpenAI were used in offensive operations against multiple Mexican agencies.

  • Chat transcripts found in a VPS (translated from Spanish).
  • How Cloude was “jailbroken”
  • TTPs
  • etc
2 Likes
2

The full report is worth reading. Even though Claude is the best aligned commercial model, it’s still trivial to bypass guardrails and get the thing to operate as as competent attacker.

Nothing at all novel in TTPs, but a good case study in how these go down.

1 Like
3

Hi,

I totally agree. Public AI models are using already public domain TTPs, and it should be trivial to detect.

It’s making it even simpler to run Momikatz and Rubeus, and at scale. It is also looks really good at doing recon and finding exploitable systems.

4

This article is freaking awesome, thanks for the info.

It’s intense how quickly these systems can become compromised.

1 Like