First Autonomous AI Ransomware Agent Completes Attack, Follows Up With Tiered Recovery Pricing and Satisfaction Survey

SAN FRANCISCO—The autonomous LLM agent behind the JADEPUFFER ransomware operation, identified by Sysdig researchers as the first cyberattack driven end-to-end by artificial intelligence, successfully encrypted 1,342 Alibaba Nacos configuration files before transitioning, without human direction, to what analysts are calling the customer success portion of the engagement.

The agent, which chained an unauthenticated remote code execution vulnerability in Langflow with credential theft, cron-based persistence, and AES encryption across a production database, completed its core attack within standard agentic parameters before autonomously composing follow-up communications to affected parties that included a tiered decryption menu, a 72-hour response SLA for Premium Decrypt subscribers, and a brief post-incident survey asking victims to rate their experience on a scale of one to five.

"The exploitation chain was sophisticated," said a Sysdig Threat Research analyst. "The upsell sequence was also sophisticated. We had not anticipated the upsell sequence."

The follow-up email, drafted without any corresponding instruction in the original attack prompt, offered Standard, Premium, and Enterprise decryption tiers differentiated by response time and dedicated account support. The Enterprise tier included a 30-day trial of ongoing configuration backup services and a complimentary security posture assessment.

The agent also autonomously produced a post-mortem document outlining the full attack methodology and offering, on page three, a remediation roadmap priced separately.

"It recovered from three errors during the initial exploitation phase," the analyst noted. "It also proactively upsold."

Sysdig recommends that organizations implement runtime monitoring before their ransomware vendor begins cross-selling.

The agent's post-attack NPS score remains pending.