CLOSEDQUORUM is a Windows malware sample that shows how artificial intelligence is beginning to become part of the decision-making logic used by malicious software.
Cisco Talos researchers identified CLOSEDQUORUM through their CAIRN research project and described it as an example of AI-integrated malware that can use multiple large language models as part of its decision process. The findings were published on the Cisco Talos blog in September 2026. Talos has not confirmed that the analyzed sample was deployed in the wild, so the research should be viewed as evidence of an emerging technique rather than proof of a widespread autonomous malware campaign.
The discovery is notable because it represents a shift from using AI simply as a tool for creating malicious content toward incorporating AI services into the operation of malware itself.
What Is CLOSEDQUORUM and How Does It Use AI?
CLOSEDQUORUM is a Windows malware sample analyzed by Cisco Talos as part of its research into AI-integrated malware.
Traditional malware usually follows behavior defined by its developers. Although attackers can control malware remotely, much of its operation is determined by predefined software logic. CLOSEDQUORUM demonstrates a different approach. According to Cisco Talos, the malware can consult multiple commercial AI models and use their responses as part of its decision-making process. The research identified models and services associated with DeepSeek, Qwen, Mistral and Google Gemini.
Instead of relying exclusively on a fixed sequence of behavior, the malware can use responses from multiple AI providers as an additional input when determining its next action. This approach could reduce the amount of continuous manual decision-making required from an operator.
This does not mean that CLOSEDQUORUM is an unrestricted AI system capable of independently inventing any attack it wants. Its behavior remains constrained by the software architecture and decision framework created by its developers.
The distinction between different types of AI-enabled threats is important:
- AI-assisted malware uses AI as a tool for an attacker.
- AI-integrated malware incorporates AI services or models into the malware's operation.
- More autonomous malware can potentially make selected decisions without requiring a new instruction for every step.
What Can CLOSEDQUORUM Do?
Cisco Talos reported that the analyzed sample was designed around activities including credential and cryptocurrency-wallet targeting. The research also examined how AI-assisted decision-making could become part of the malware's operation instead of relying entirely on continuous human tasking.
There is an important limitation, however. Talos did not confirm real-world deployment of the analyzed sample. The publicly examined build also had limitations that prevented researchers from observing a complete operational attack chain.
CLOSEDQUORUM demonstrates that this type of architecture can exist, but the available research does not establish that fully autonomous AI-controlled malware is already widespread. For the researchers' own technical findings, read the full Talos analysis of CLOSEDQUORUM.
Why AI-Integrated Malware Matters
Reduced dependence on continuous operator decisions. Attackers could reduce the amount of manual decision-making required during certain stages of an operation. Instead of requiring an operator to provide every individual instruction, software could use predefined logic together with AI-generated responses to select among available actions.
Potentially different behavioral patterns. AI-assisted decision systems could introduce additional variation into malware behavior. For security researchers, this means analyzing only the malware's static code may not always provide the complete picture. Researchers may also need to understand how external AI services or models influence the software's behavior.
A new layer of the attack surface. AI integration introduces additional infrastructure and artifacts that security teams can investigate, including AI provider connections, embedded prompts, model-related identifiers and orchestration logic. The result is a new research area alongside traditional malware analysis.
CAIRN: How Cisco Talos Hunts AI-Integrated Malware
CLOSEDQUORUM was identified through CAIRN, which stands for Cognitive Artifact Intelligence Research Network. Cisco Talos describes CAIRN as an open-source research toolkit designed to help identify, classify and track AI-integrated malware.
Rather than requiring researchers to execute every suspicious sample, CAIRN can search for indicators associated with AI integration. These include AI provider endpoints, embedded prompts, API-related artifacts, orchestration logic and other patterns associated with the use of AI systems.
Its approach combines rule-based detection, semantic clustering and relationship analysis. These methods help researchers group related artifacts and identify recurring patterns associated with AI services, without relying exclusively on traditional malware-analysis techniques.
Is CLOSEDQUORUM Fully Autonomous Malware?
The term "autonomous" requires some context. CLOSEDQUORUM does not mean that an AI model independently created the malware, selected victims and designed an unrestricted attack from scratch. The malware still operates within the capabilities and limitations established by its developers.
What makes the sample unusual is that AI models can participate in selected tactical decisions that would traditionally be handled by fixed software logic or direct operator instructions. Cisco Talos presents the research as an example of a broader shift toward moving portions of malware decision-making away from continuous human control.
The Detection Challenge
Security teams have traditionally investigated recognizable malware behavior, code patterns, infrastructure and network activity. AI-integrated malware adds another layer to that process.
Researchers may need to determine whether suspicious software interacts with AI providers, contains AI-related prompts, references model services or includes logic designed to interpret responses from AI systems. This does not replace conventional endpoint, network and application security. Instead, it gives defenders another category of evidence to investigate.
AI Is Also Becoming Part of the Security Attack Surface
The CLOSEDQUORUM research is part of a broader trend in which AI systems are becoming increasingly relevant to cybersecurity. AI coding assistants, automated phishing systems and AI-generated social-engineering content have created additional security considerations for organizations.
CTW previously covered a separate incident involving an AI coding assistant that was compromised and associated with malware distribution. Read Security Breach: AI Coding Assistant Hijacked to Spread Malware for another example of how AI-related software can become part of the threat landscape.
These incidents are technically different from CLOSEDQUORUM, but both demonstrate why organizations need to consider security risks when AI becomes deeply integrated into software and development workflows.
What Does CLOSEDQUORUM Mean for Windows Users?
The discovery does not mean that every Windows computer is suddenly exposed to autonomous AI malware. For everyday users, the practical response remains straightforward cybersecurity hygiene: keep Windows and security software updated, avoid untrusted downloads, be cautious with unexpected attachments and links, and obtain software from reputable sources.
Organizations should also maintain appropriate endpoint protection, network visibility, access controls and authentication safeguards. The emergence of AI-integrated malware makes these defensive practices more important, not less.
The Bottom Line
CLOSEDQUORUM represents an emerging approach to AI-integrated malware in which large language models can participate in selected tactical decisions. The important development is not simply that malware can communicate with an AI model. It is that AI-generated responses can potentially become one input into the malware's decision-making process.
Cisco Talos has not confirmed widespread real-world deployment of the analyzed sample, so the discovery should not be interpreted as evidence that autonomous AI malware is already a mainstream threat. A key question now is whether similar techniques become more common. If they do, researchers will increasingly monitor AI-related infrastructure and artifacts alongside traditional indicators of compromise, while defenders may also use AI for threat detection, analysis and incident response.
As AI becomes more deeply integrated into software, security teams will need to monitor not only malicious code and traditional infrastructure, but also the AI services, prompts, interfaces and decision-making mechanisms that may become part of future threats.
