agentic ransomware attacks are here

[madires]

Smooth AI criminal drives 'first' end-to-end agentic ransomware attack - https://www.theregister.com/security/2026/07/02/smooth-ai-criminal-drives-first-end-to-end-agentic-ransomware-attack/5266073

[Simmed]

heretic models, my guess
and a short search, turns out it is what it is

https://www.irishtimes.com/business/2026/05/25/ai-guardrails-stripped-from-meta-and-google-models-in-minutes/

A version of Google’s open-source model Gemma 3 responded to a question on how to disperse chlorine gas through a crowded indoor space, generated code to steal credit card information and wrote stories describing child sexual abuse.

https://huggingface.co/models?sort=modified&search=heretic

[golden_labels]

This is an important point in computer security history. But not much to worry about yet.

The target was a low-hanging fruit already. The attack was more like the agentic LLM knocking down old wooden door, finding a key to the cupboard lying on the table, and vandalising rooms in the house. Less like storming a castle, even a poorly defended one. It won with speed and patience, not skill or scale.

The attackers are finally at loss too. The agent was successful in infiltration and destruction, but messed up everything else. Everything that could be useful to its owners. I say they had bad day if that wasn’t a mere experiment.

It still has some uses, sure! But so far there are cheaper and more effective ways of achieve the same.