CLOSEDQUORUM reportedly lets AI models choose post-compromise actions
SecurityUnverifiedUpdated:
A reported new Windows malware sample, CLOSEDQUORUM, shifts the choice of its next predefined post-compromise action from an ongoing human operator to a vote among up to four AI models.
Four AI models vote on predefined actions
The Register reported that Cisco Talos identified a Go-based Windows malware called CLOSEDQUORUM that can query Google Gemini, DeepSeek, Qwen, and Mistral.
According to the report, the models vote among actions already built into the malware, such as credential and cryptocurrency-wallet theft, code injection, and persistence. The report says the malware can act after deployment without continued commands from a human operator, but no in-the-wild deployment had been observed.
Post-compromise actions in malware are commonly directed by a human operator or by fixed automated logic.
CLOSEDQUORUM reportedly adds an LLM-voting mechanism that selects among predefined post-compromise actions after deployment without continued human commands.
No deployment in the wild observed
If the reported design is accurate, it moves the selection of a malware's next action into an AI-model voting process while retaining a bounded set of executable modules. Talos characterized it as the first publicly documented Windows implant using this approach for command-and-control.
Sources
- The RegisterWindows CLOSEDQUORUM malware uses AI models to autonomously select post-compromise actionsRead source →