# New ‘CLOSEDQUORUM’ Windows Malware Lets AI Models Decide How to Steal Data

*Wednesday, September 23, 2026 at 4:08 PM UTC — Hamer Intelligence Services Desk*

**Published**: 2026-09-23T16:08:22.138Z (2h ago)
**Category**: cyber | **Region**: Global
**Importance**: 7/10
**Sources**: OSINT
**Permalink**: https://hamerintel.com/data/articles/18650.md
**Source**: https://hamerintel.com/summaries

---

**Deck**: A Windows malware strain dubbed CLOSEDQUORUM uses up to four built-in AI models to vote on its next move instead of waiting for attacker commands, making it harder to disrupt while it hunts for credentials, crypto wallets, and code injection opportunities.

A new Windows malware family is testing how far attackers can go by letting artificial intelligence make decisions on infected machines.

The strain, called CLOSEDQUORUM, breaks with the usual pattern in which malware calls back to a remote command-and-control server for instructions. Instead, it carries up to four embedded AI models, including DeepSeek and Gemini, and lets them choose what to do next.

According to analysis cited by The Hacker News, these models effectively vote on actions such as stealing credentials and wallet data, injecting code, or remaining installed but inactive to preserve access. Because CLOSEDQUORUM can operate without constant external guidance, defenders have fewer chances to spot or cut off suspicious network traffic.

The immediate targets are valuable pieces of information stored on everyday systems. The malware can go after saved passwords, authentication tokens, and data from cryptocurrency wallets, putting both personal finances and corporate access points at risk.

For organisations, CLOSEDQUORUM underscores a shift from centrally controlled botnets to more autonomous malicious software. Security tools tuned to watch for unusual outbound connections may miss a threat that does most of its thinking and planning locally, relying on built-in models to adapt to each environment it lands in.

The case also shows how quickly AI capabilities are being pulled into offensive cyber tools. Rather than just using AI to write malicious code, CLOSEDQUORUM uses it to orchestrate an infection in real time, choosing tactics based on what it finds on the victim system.

Defenders will need to watch for behaviour that suggests on-device model execution linked to suspicious activity, tighten controls around credential storage and wallet usage on general-purpose PCs, and update detection rules for malware that doesn’t behave like traditional remote-controlled threats.

Key questions now are whether CLOSEDQUORUM appears widely in real-world attacks, whether copycat malware families adopt similar embedded-model designs, and how software and security vendors respond to the idea of AI-powered decision-making inside malicious code.