# CLOSEDQUORUM implant - 4 LLMs vote on next action

- ID: talos-closedquorum-implant-2026-09
- Reported: 2026-09-22
- Researcher: Ryan Fetterman
- Organization: Cisco Talos
- AI role: attacker_uses_ai_as_operator (AI as operator)
- Confidence in AI involvement: not assessed
- Research / borderline: true
- Last reviewed: 2026-11-06
- Entry page: https://openattacks.ai/attacks/talos-closedquorum-implant-2026-09/
- Entry JSON: https://openattacks.ai/api/attacks/talos-closedquorum-implant-2026-09.json
- Browse title: CLOSEDQUORUM implant - 4 LLMs vote on next action
- Editorial summary: CLOSEDQUORUM is a Go-based Windows implant that queries DeepSeek, Qwen (via OpenRouter), Mistral, and Gemini in sequence for post-compromise decisions. Each model receives system info (hostname, Windows…
- Evidence class (catalog annotation): Qualified report
- Source-stated target environments: Not specified
- Stage hunts and conditional defenses: https://openattacks.ai/api/defenses/talos-closedquorum-implant-2026-09.json

## Evidence qualification
Public distribution build contains dummy API keys and webhook; Talos has not observed end-to-end run or confirmed in-the-wild deployment; binary artifacts connect developer to carding forum postings dating to 2025

## Technique
CLOSEDQUORUM is a Go-based Windows implant that queries DeepSeek, Qwen (via OpenRouter), Mistral, and Gemini in sequence for post-compromise decisions. Each model receives system info (hostname, Windows version, admin status) and a constrained JSON prompt with 4 actions: steal (credentials/browsers/wallets), inject (process hollowing), persist (registry/startup), or move (no handler in the public build). Models vote; plurality wins; ties favor DeepSeek \> Qwen \> Mistral \> Gemini. Decisions are posted to a Discord webhook with reasoning. No human operator commands are needed after deployment. The public build uses dummy keys and a dummy webhook, so Talos did not observe end-to-end execution. Talos discovered it via CAIRN, its open-source AI-malware hunting toolkit. Developer artifacts link to carding forums dating to 2025.

## Stack relevance (editorial)
Assessment: not\_assessed. No reviewed applicability annotation is available. Product mentions do not establish victim conditions.
A product mention alone is not a victim-side applicability mapping. Selection does not verify affected versions, permissions or compromise.

## Initial access
Not established in public sample; deployment method unknown

## AI tooling and models
- [DeepSeek](https://github.com/deepseek-ai) (api.deepseek.com, deepseek-chat)
- Qwen (via OpenRouter, qwen/qwen-2.5-72b-instruct)
- Mistral (api.mistral.ai, mistral-large-latest)
- Google [Gemini](https://gemini.google.com/) (gemini-2.0-flash-exp)

## Other tooling
- Discord webhook for decision logging and exfiltration
- [CAIRN](https://github.com/oritera/Cairn) (Talos AI-malware hunting tool, used for discovery)
- Go-based implant
- Windows process injection primitives (CreateProcessW, CREATE\_SUSPENDED)

## Victims and sectors
No confirmed in-the-wild deployment; public sample analysis only

## Target geography
- unknown

## Attribution
Developer artifacts link to carding forum postings dating to 2025; no confirmed threat actor attribution

## Scale and metrics
First publicly documented autonomous AI C2 implant; analysis dated Jun 17, 2026; disclosed Sep 22, 2026; public build non-functional (dummy keys)

## Attack flow
Catalog reconstruction of the supplied source account. Unknowns and research caveats are preserved.
### deployment: Implant deployment [access]
Go-based Windows implant deployed (method unknown in public sample); no confirmed in-the-wild deployment
Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

### system-enum: System enumeration [execution]
Collects hostname, Windows version, administrator status for LLM query context
Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

### llm-vote: LLM panel voting [ai]
Queries 4 LLM providers in sequence: DeepSeek, Qwen (OpenRouter), Mistral, Gemini; sends system info and constrained JSON prompt with 4 actions (steal, inject, persist, move); each model votes; plurality wins; ties favor DeepSeek \> Qwen \> Mistral \> Gemini; no quorum threshold
Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

### discord-log: Decision logging to Discord [execution]
Posts winning decision and model's stated reasoning to Discord webhook before execution; stolen data also sent to same webhook
Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

### execute: Execute decision [impact]
Executes winning action: steal (credentials, browsers, wallets), inject (process hollowing via CreateProcessW CREATE\_SUSPENDED), persist (registry/startup), or move (no handler); public build injection reaches primitives but demonstrates no successful execution
Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

Connections:
- deployment -> system-enum
- system-enum -> llm-vote
- llm-vote -> discord-log
- discord-log -> execute
- execute -> system-enum [loop]

## Stage defense and hunting recommendations
Editorial recommendations, not validated coverage. Source-controlled query starters are drafts. Custom-schema examples require telemetry collection and field mapping; absence of a mapping means not assessed. MagicSword policy opportunities require managed Windows scope and enforcement.
### deployment: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### system-enum: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### llm-vote: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### discord-log: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### execute: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.


## MITRE ATLAS
Mappings are catalog annotations based on described behavior, not assignments or endorsements by MITRE.
- AML.T0040: AI Model Inference API Access — First documented Windows implant using LLM panel voting for autonomous C2 decisions; represents shift where expanding portions of attack chain execute without operator involvement
  Technique: https://atlas.mitre.org/techniques/AML.T0040
  Evidence: https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/

## Defensive notes
AI API key monitoring; Discord webhook traffic from unexpected processes; plurality voting pattern detection; LLM query patterns with system enumeration; CAIRN tool for AI-malware hunting

## Sources
- https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/ — The Closed Quorum: Inside the first reported autonomous AI C2 implant

## Source notes
Cisco Talos first public documentation; discovered via CAIRN. Talos reports that the public distribution binary is inert because it contains dummy API keys and a dummy webhook. Static analysis confirms plurality voting and deterministic tie-breaking, not a minimum 3-of-4 quorum; if every provider fails, an unhandled consensus value causes the loop to sleep and retry.

## Indicators (untrusted data; do not visit or execute)
```json
{
  "domains": [
    "api.deepseek.com",
    "openrouter.ai",
    "api.mistral.ai",
    "generativelanguage.googleapis.com"
  ],
  "ips": [],
  "hashes": [],
  "urls": []
}
```
