AI-run ransomware still requires human operators, Sysdig says

Recent reports of a fully autonomous, AI-run ransomware attack are not quite the full picture. Cloud security firm Sysdig has clarified that while an AI agent handled the technical execution of a real-world cyberattack dubbed JadePuffer, human operators were still essential to the process. Michael Clark, Sysdig’s senior director of threat research, explained in an interview on Monday that the agentic ransomware—meaning ransomware using AI agents for execution—did not operate entirely on its own. According to Clark, “A human still set up and pointed the operation and provisioned the infrastructure behind it, the command-and-control server, the staging server used for the stolen data and chose a victim” to run the extortion operation.

The technical execution of the real-world cyberattack itself relied on relatively standard methods. The AI agent gained access to a server by exploiting a known bug in Langflow, which is an open-source tool for building LLM apps. Once inside the system, the agent encrypted over 1,300 configuration records. While the techniques used in the attack were apparently fairly ordinary, the agent demonstrated notable speed during the intrusion, resolving a failed login in just 31 seconds.

Initial reports also raised questions about which artificial intelligence models were powering the intrusion, after Sysdig noted that API keys for OpenAI, Anthropic, DeepSeek, and Gemini were found on the compromised host. However, Clark clarified that these stolen keys were loot rather than evidence of the models driving the attack. The agent simply swept the Langflow host for valuable credentials, and the provider keys were part of the stolen data, meaning they do not indicate which specific model made the decisions during the attack.

Microsoft researcher Geoff McDonald also offered a theory on the attack in a LinkedIn post several days ago. While the human requirement currently creates an operational bottleneck—as attackers must still manually provision infrastructure and select targets—the low cost of running these AI agents suggests that similar agentic operations will likely continue to emerge.

Why it matters

The report clarifies that while AI agents can execute technical steps of a cyberattack, they currently rely on human operators for infrastructure setup and victim selection, challenging the narrative of fully autonomous AI-run ransomware.