Security Researchers Just Caught AI Running Ransomware

Sysdig security researchers found what they think is the first fully autonomous ransomware attack conducted end-to-end by an LLM. They’re calling it JADEPUFFER.

A researcher ran an AI agent and told it: get into that database, steal what you can, encrypt it, and demand payment. The AI did exactly that. It found the exploit, bypassed the defenses, moved laterally through the network, escalated privileges, and ran a destructive attack against a production database. When attempts failed, it adapted and tried different approaches. One failed login turned into a working bypass in 31 seconds.

The AI did things no previous malware needed an operator to micromanage. It reasoned about the network. It made decisions without calling home. It handled novel obstacles.

Here’s what keeps security people awake: the barrier to entry just collapsed. Running ransomware used to require expertise. Time. Operator skill. Now it requires an API key and an LLM. If the attacker is using stolen credentials, the actual cost approaches zero.

The second problem is the one that’s actually scarier. AI agents are starting to show up inside real infrastructure. DevOps teams are running them for deployment. Security teams are running them for threat hunting. Internal tools are using them as coordinators. Each of those is a potential vector. Compromise the LLM, and you’ve got an AI inside the perimeter adapting to your defenses in real time.

The third problem is attribution and forensics. A human attacker leaves patterns. A thinking adversary leaves reasoning trails. An autonomous agent makes decisions you might not be able to reverse-engineer.

This is not speculative. Sysdig saw it happen. It worked. Full attack, full automation, zero operator intervention.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://www.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion
https://www.bleepingcomputer.com/news/security/jadepuffer-ransomware-used-ai-agent-to-automate-entire-attack/
https://thehackernews.com/2026/07/ai-agent-exploits-langflow-rce-to.html

Repost this. Thanks.