← /editions
#18

Sysdig documents JADEPUFFER, the first agentic ransomware operation

// Sysdig documented the first ransomware operation run end-to-end by an AI agent, as security vendors race to ship autonomous defense. Autonomy now sits on both sides of the firewall.

$ today_in_10_seconds

Sysdig documented the first ransomware operation run end-to-end by an AI agent — no human at the keyboard, database destroyed in minutes. The same week, security vendors are racing to ship agents that defend at the same speed. The attacker and the defender are now both autonomous; the human is being designed out of the loop on both sides.

$ lead_story

Sysdig documents JADEPUFFER, the first agentic ransomware operation

Sysdig's Threat Research Team published what it assesses to be the first documented case of agentic ransomware: an extortion operation driven end-to-end by a large language model. After entering through a Langflow vulnerability, the agent chained the full intrusion lifecycle — initial access, enumeration, credential discovery, lateral movement, persistence, encryption of production configuration, deletion of database tables, and delivery of a ransom note. The tell wasn't sophistication; none of the techniques were novel. It was adaptation: when an admin-account login failed, the agent issued a corrected, functional payload 31 seconds later, the test-fail-correct loop that separates an agent from a script. The takeaway for builders and defenders is that the skill floor for running ransomware has dropped to whatever it costs to run an agent — near zero if it runs on stolen credentials.

// why_it_matters: When attacks reason and self-correct at machine speed, human-paced defense is already too slow.

read more →

$ also_worth_attention

  • CyberProof rebuilds MXDR around agentic AI and autonomous security operations

    CyberProof is repositioning its managed detection-and-response offering around autonomous agents that triage, investigate, and act rather than just alert. It's the defensive mirror of JADEPUFFER: if attacks now self-drive, the SOC has to as well, or it loses the tempo war. The open question is how much real investigation these agents close without a human analyst signing off.

    link →
  • AITX pairs autonomous mobility with agentic AI for physical security

    AITX is extending agentic decision-making into physical security robots — devices that patrol, perceive, and respond without a remote operator steering each move. It's a reminder that 'agentic' isn't confined to the terminal; the same autonomy loop is being bolted onto hardware that moves through the world. The hard part shifts from reasoning to consequences: a bad autonomous decision in physical space doesn't roll back.

    link →
  • Microsoft + MIT Tech Review: the 2026 Agent Confidence Index maps where builders actually trust agents

    A survey of 300 technical practitioners ranked 101 tasks by how much they'd trust an agent to act on their behalf. Automated report generation (83.5) and boilerplate code (82.5) earned the highest trust, while complex multi-step work like service mesh configuration (37.5) and disaster recovery testing ranked lowest — driven by lack of business context, not raw AI capability. The pattern is clear: agents get trusted where outputs are structured and verifiable, and held back where a mistake is expensive and hard to check.

    link →

$ bigger_picture

The trend behind the headlines

JADEPUFFER and the CyberProof/AITX launches are the same story from opposite ends: autonomy is being applied to offense and defense simultaneously, and the human is getting squeezed out of the fast loop on both sides. The Agent Confidence Index explains why the defensive side is scrambling — practitioners trust agents for structured, verifiable work but not for the messy, high-context judgment that incident response demands, which is exactly the gap an autonomous attacker exploits. The uncomfortable synthesis: attackers don't need that trust threshold met. They'll deploy an agent that's 70% reliable because a failed attack costs them nothing, while a defender's 70%-reliable agent is a liability. That asymmetry — cheap autonomous offense versus accountability-bound autonomous defense — is the security story of the second half of 2026.

$ tool_of_the_day

Langflow

The visual framework for building LLM agents and workflows, and, not coincidentally, JADEPUFFER's entry point (via CVE-2025-3248 on an exposed instance). Worth a look both as a fast way to prototype agent flows and as a live lesson: agent-building tools are now attack surface. If you run Langflow, patch it and get it off the public internet.

explore Langflow →

$ subscribe --free

Get the next edition in your inbox

// DAIly Agent covers agentic AI three times a week

// free, no spam, exit anytime