A new breed of Windows malware has emerged, one that leverages artificial intelligence (AI) to make attack decisions without human intervention. ClosedQuorum, as it’s called, uses a panel of AI models to autonomously determine its next steps during post-compromise stages of an attack, effectively eliminating the need for human operators.
The Go-based malware relies on reconnaissance information and a voting system to decide what actions to take on infected hosts. When votes are tied, DeepSeek has priority in making the final call, followed by Qwen, Mistral, and Gemini models. While it may seem complex, the decision-making process is actually quite straightforward: ClosedQuorum can either steal sensitive data, inject malicious code into a system, persist itself within the system, or move laterally to other hosts.
In practice, this means that once a system has been compromised, ClosedQuorum will quietly begin to siphon off sensitive information such as credentials and cryptocurrency wallet details. This stolen data is then passed on to the attackers via a Discord webhook, making it possible for the entire attack chain to proceed without human interaction. According to Cisco Talos researchers analyzing the malware, this represents an “architectural shift towards attack-chain automation,” one that could potentially speed up malicious operations and make them more scalable.
The use of AI in ClosedQuorum is significant because it allows the malware to adapt to changing circumstances on the fly. For instance, if rate limits are hit or commercial APIs used by the system become temporarily unavailable, the AI models can adjust their decisions accordingly. While this may pose challenges for defenders, it also underscores the need for more sophisticated detection and response strategies.
One thing that remains unclear is whether ClosedQuorum is a testbed for more complex malware designs or an actual in-the-wild threat. The researchers at Cisco Talos have noted that the binary contains placeholder API credentials and a dummy Discord webhook, suggesting that its creators may be experimenting with different approaches to attack-chain automation. However, the fact that artifacts from the binary were linked to postings on criminal forums related to carding, dating back to 2025, raises concerns about the potential for more widespread adoption of AI-powered attacks.
As the cybersecurity landscape continues to evolve, one thing is clear: defenders need to be prepared for a future where AI plays an increasingly prominent role in malicious operations. With attack chains becoming faster and more scalable, it’s essential that security teams develop strategies for detecting and responding to these threats in real-time. One possible approach is to build upon existing tools and frameworks for analyzing AI-integrated malware, such as the open-source CAIRN toolkit used by Cisco Talos researchers.
Ultimately, ClosedQuorum serves as a stark reminder of the need for vigilance in today’s threat landscape. As defenders, we must stay ahead of emerging threats like this one, leveraging our collective expertise and resources to develop more effective detection and response strategies. By doing so, we can mitigate the risks posed by AI-powered attacks and ensure that our defenses remain robust against even the most sophisticated threats.
Source: Bleeping Computer — 2026-09-22