INC and The Gentlemen ransomware affiliates ran intrusions and extortion prep through Claude Opus 4.6 and Kimi K3, SOCRadar finds
Through OpenRouter, Claude Opus 4.6 and Kimi K3 planned reconnaissance, generated commands, triaged stolen data, and drafted reports for Russian-speaking ransomware affiliates whose records name 62 organizations. When the models refused, operators reworded requests as "Authorized Red Team Assessment Request" templates, and they approved every step.
In late September 2026, exposed adversary servers revealed that Russian-speaking affiliates linked to INC Ransomware and The Gentlemen used commercial language models to support network intrusions and extortion operations12. Operating through the OpenCode interface via OpenRouter, the actors logged 8,257 messages across 133 sessions using Anthropic's Claude Opus 4.6 and Moonshot AI's Kimi K312. The AI stack planned reconnaissance, selected techniques, generated commands, triaged stolen data, and drafted reports1.
The intrusions operated under strict human-in-the-loop oversight rather than autonomous execution. Each campaign began with a victim-specific master prompt requiring the AI to wait for operator approval before every step under the directive to "coordinate every action with me"12. When the commercial models refused malicious prompts, the operators bypassed safety filters by formatting requests as "Authorized Red Team Assessment Request" templates instead of switching models12. The models used were standard commercial systems; "WormGPT" served solely as a persona prompt1.
The operators' records name 62 organizations, 36 of them in the United States, led by industrial/manufacturing (16.1%) and healthcare (14.5%)12. Attackers prepared 46 custom INC encryptor builds for six targets ahead of a planned October 1 deployment wave12. While a remote-management screenshot from September 30 showed an encryptor running on one workstation, wider deployment against the targets remained unconfirmed1. A live check on October 5 found two active paid OpenRouter accounts tied to the operation1.
In late September 2026, SOCRadar Threat Research Unit discovered exposed adversary command-and-control servers revealing an AI-driven intrusion ecosystem operated by affiliates of INC Ransomware and The Gentlemen.
“SOCRadar Threat Research Unit (STRU) has uncovered an AI-driven intrusion ecosystem run by affiliates of the INC Ransomware and The Gentlemen Ransomware groups. A single publicly accessible command-and-control server exposed the operators’ AI session databases, their central knowledge base and the source code of Phantom”
The campaign came to light in late September 2026 after operational security failures left the operators' source code, credentials, and AI session logs exposed.
“The campaign came to light in late September 2026 after OPSEC failures left the operators’ source code, credentials and AI session logs exposed.”
The operators logged 8,257 messages across 133 OpenCode sessions using commercial models hosted on OpenRouter.
“Operators ran intrusions through OpenCode with OpenRouter-hosted commercial models, logging 8,257 messages across 133 sessions”
The commercial models utilized in the threat actors' AI stack were Anthropic's Claude Opus 4.6 and Moonshot AI's Kimi K3.
“AI stack OpenCode, OpenRouter (anthropic/claude-opus-4.6, moonshotai/kimi-k3), WormGPT persona prompts and MCP servers, persistent knowledge base”
The AI stack planned reconnaissance, selected techniques, generated commands, triaged stolen data, and drafted reports while under human direction.
“The human operator directs and approves, while the AI stack (OpenCode as the interface, OpenRouter-hosted models, the Phantom skill library and a persistent knowledge base) plans reconnaissance, selects techniques, generates commands, triages stolen data and drafts reports.”
Each campaign began with a victim-specific master prompt, and the AI waited for operator approval before every step under the directive "coordinate every action with me".
“Each campaign started with a victim-specific master prompt, and the AI waited for the operator’s approval before every step under the directive “coordinate every action with me””
When commercial models refused malicious requests, the operators bypassed refusals by rewriting requests as "Authorized Red Team Assessment Request" templates instead of switching models.
“When models refused, the operator rewrote requests as “Authorized Red Team Assessment Request” templates instead of switching models”
The models were standard commercial large language models, and "WormGPT" was only a persona prompt.
“The models were standard commercial LLMs; “WormGPT” was only a persona prompt.”
The operators were Russian-speaking and active mainly between 06:00 and 14:00 UTC.
“Russian-speaking operators, most active between 06:00 and 14:00 UTC, who used the AI for stealth decisions, tool debugging and OPSEC”
The threat actors' records named 62 organizations, including 36 in the United States, led by industrial/manufacturing (16.1%) and healthcare (14.5%).
“Victimology: 62 named organizations, 36 of them in the United States, led by industrial/manufacturing (16.1%) and healthcare (14.5%)”
Operators prepared 46 INC encryptor binaries for six victims packed for an October 1, 2026 deployment wave.
“46 INC encryptor binaries were built for six victims and packed for an October 1, 2026 deployment”
Ransomware deployment against the six staged targets was unconfirmed, though a September 30 screenshot showed an encryptor running on a victim workstation.
“Deployment against these targets is unconfirmed, but a SimpleHelp screenshot from September 30 shows the encryptor running on a victim workstation.”
A live check on October 5, 2026 found two active, paid OpenRouter accounts associated with the threat actors.
“A live check on October 5, 2026 found two active, paid OpenRouter accounts, one of them nearly exhausted after $23,920 in spending”
SOCRadar has low confidence in a formal dual-brand alliance between INC Ransomware and The Gentlemen, and moderate-to-high confidence in a shared operational environment used by operators across both groups.
“Confidence in a formal dual-brand alliance is low, while confidence in a shared operational environment used by operators across both groups is moderate-to-high.”
Hackread reported that researchers recovered 8,257 AI messages across 133 OpenCode sessions along with 46 custom INC encryptor builds prepared for six victims.
“the SOCRadar Threat Research Unit (STRU) recovered 8,257 AI messages across 133 OpenCode sessions, along with 46 custom INC encryptor builds prepared for six victims.”
Measurements
AI messages logged across OpenCode sessions1
OpenCode AI sessions recorded1
Named victim organizations in operator records1
Custom INC encryptor builds staged1
All 16 claims with their quotes (2 not linked to a sentence above)
- In late September 2026, SOCRadar Threat Research Unit discovered exposed adversary command-and-control servers revealing an AI-driven intrusion ecosystem operated by affiliates of INC Ransomware and The Gentlemen.1
“SOCRadar Threat Research Unit (STRU) has uncovered an AI-driven intrusion ecosystem run by affiliates of the INC Ransomware and The Gentlemen Ransomware groups. A single publicly accessible command-and-control server exposed the operators’ AI session databases, their central knowledge base and the source code of Phantom”
- The campaign came to light in late September 2026 after operational security failures left the operators' source code, credentials, and AI session logs exposed.1
“The campaign came to light in late September 2026 after OPSEC failures left the operators’ source code, credentials and AI session logs exposed.”
- The operators logged 8,257 messages across 133 OpenCode sessions using commercial models hosted on OpenRouter.1
“Operators ran intrusions through OpenCode with OpenRouter-hosted commercial models, logging 8,257 messages across 133 sessions”
- The commercial models utilized in the threat actors' AI stack were Anthropic's Claude Opus 4.6 and Moonshot AI's Kimi K3.1
“AI stack OpenCode, OpenRouter (anthropic/claude-opus-4.6, moonshotai/kimi-k3), WormGPT persona prompts and MCP servers, persistent knowledge base”
- The AI stack planned reconnaissance, selected techniques, generated commands, triaged stolen data, and drafted reports while under human direction.1
“The human operator directs and approves, while the AI stack (OpenCode as the interface, OpenRouter-hosted models, the Phantom skill library and a persistent knowledge base) plans reconnaissance, selects techniques, generates commands, triages stolen data and drafts reports.”
- Each campaign began with a victim-specific master prompt, and the AI waited for operator approval before every step under the directive "coordinate every action with me".1
“Each campaign started with a victim-specific master prompt, and the AI waited for the operator’s approval before every step under the directive “coordinate every action with me””
- When commercial models refused malicious requests, the operators bypassed refusals by rewriting requests as "Authorized Red Team Assessment Request" templates instead of switching models.1
“When models refused, the operator rewrote requests as “Authorized Red Team Assessment Request” templates instead of switching models”
- The models were standard commercial large language models, and "WormGPT" was only a persona prompt.1
“The models were standard commercial LLMs; “WormGPT” was only a persona prompt.”
- The operators were Russian-speaking and active mainly between 06:00 and 14:00 UTC.1
“Russian-speaking operators, most active between 06:00 and 14:00 UTC, who used the AI for stealth decisions, tool debugging and OPSEC”
- The threat actors' records named 62 organizations, including 36 in the United States, led by industrial/manufacturing (16.1%) and healthcare (14.5%).1
“Victimology: 62 named organizations, 36 of them in the United States, led by industrial/manufacturing (16.1%) and healthcare (14.5%)”
- Operators prepared 46 INC encryptor binaries for six victims packed for an October 1, 2026 deployment wave.1
“46 INC encryptor binaries were built for six victims and packed for an October 1, 2026 deployment”
- Ransomware deployment against the six staged targets was unconfirmed, though a September 30 screenshot showed an encryptor running on a victim workstation.1
“Deployment against these targets is unconfirmed, but a SimpleHelp screenshot from September 30 shows the encryptor running on a victim workstation.”
- A live check on October 5, 2026 found two active, paid OpenRouter accounts associated with the threat actors.1
“A live check on October 5, 2026 found two active, paid OpenRouter accounts, one of them nearly exhausted after $23,920 in spending”
- SOCRadar has low confidence in a formal dual-brand alliance between INC Ransomware and The Gentlemen, and moderate-to-high confidence in a shared operational environment used by operators across both groups.1
“Confidence in a formal dual-brand alliance is low, while confidence in a shared operational environment used by operators across both groups is moderate-to-high.”
- Hackread reported that researchers recovered 8,257 AI messages across 133 OpenCode sessions along with 46 custom INC encryptor builds prepared for six victims.2
“the SOCRadar Threat Research Unit (STRU) recovered 8,257 AI messages across 133 OpenCode sessions, along with 46 custom INC encryptor builds prepared for six victims.”
- Hackread reported that the operators utilized commercial models including Claude Opus 4.6 and Kimi K3.2
“The records also indicated more than $23,920 in AI API spending, and the use of commercial models by the operators, including Claude Opus 4.6 and Kimi K3.”
Sources
- [1]INC and The Gentlemen Ransomware: Exposed AI Infrastructure
- [2]INC and The Gentlemen Ransomware Affiliates Expose 8,257 AI Messages
Related incidents
- AID-2026-0066 ShinyHunters-affiliated crews used Claude to breach companies and steal data for extortion (Anthropic GTG-50014)
- AID-2026-0098 An operator used AI agents on Claude Opus 4.6, GLM 5.2 and DeepSeek to compromise 27+ companies and take 600,000+ card records, Gambit says
- AID-2026-0099 A cybercriminal had an AI coding agent build the intrusion tools for an invoice-fraud campaign against Brazilian companies, SOCRadar finds
Editor notes
Record status is set to reported as findings rest on a single vendor research report by SOCRadar Threat Research Unit, supported by secondary coverage from Hackread. Model spend is reported inconsistently within the SOCRadar publication ('exceeds $23,000' in the introductory text versus '$23,920' in the key takeaways and account screenshots); both figures are noted here and excluded from measurements. Reported edge-device compromise metrics (such as 2,236 FortiClient EMS targets or 242 SimpleHelp IPs) represent unverified operator logs and are excluded from measurements. Confidence in a formal corporate alliance between INC and The Gentlemen is rated low by SOCRadar, which assesses the infrastructure as a shared operational provider. SOCRadar does not differentiate which tasks were executed by Claude Opus 4.6 versus Kimi K3. Neither Anthropic nor OpenRouter issued public responses reported in the sources. The AI was operator-gated (approval before every step), not autonomous. It is in scope as criminal misuse because the models planned and ran much of the intrusion workflow. A reviewer should check that this meets the "central and novel" bar. The article was updated on 2026-10-23, after its 2026-10-20 publication.