ClosedQuorum reportedly delegates attack decisions to AI models
AI securityUnverifiedUpdated:
A newly documented Windows malware called ClosedQuorum reportedly lets several AI models select its next post-compromise action on an infected host rather than waiting for a human operator's command.
AI selects a next action from reconnaissance data
According to the report, the Go-based malware uses reconnaissance data and models from Google Gemini, DeepSeek, Qwen, and Mistral to vote on a next action. The available choices include credential theft, code injection, persistence, and a lateral-movement option that the analyzed build cannot execute. Cisco Talos says it has no confirmation that ClosedQuorum has been deployed in the wild.
Cisco Talos describes this as the first publicly documented Windows implant to delegate tactical command-and-control decisions to a panel of AI models.
ClosedQuorum reportedly shifts post-compromise decision-making from live human operator commands to a vote among multiple AI models, within a predefined set of actions.
Lateral movement absent from the analyzed build
If the analysis is accurate, the malware architecture would automate tactical choices after compromise while constraining those choices to a preset menu. The report also identifies operational dependencies, including API availability, rate limits, and malformed model output.
Sources
- BleepingComputerNew ClosedQuorum Windows malware uses AI for attack decisionsRead source →