Threat Advisories:
New Report Critical Threat Research : The Iranian Cyber War Intensifies! Download the Report

Gemini on Payroll: A Solo Actor Outsources a Five-Year Influence Op to AI

Amber | Attack Report
Download PDF

Summary

“Patriot Bait” is a newly uncovered AI-powered influence operation in which a solo Russian-speaking threat actor tracked as bandcampro ran a MAGA-themed Telegram channel of approximately 17,000 subscribers from February 6, 2021, pivoting in September 2025 to AI-automated content generation, cryptocurrency fraud, and credential theft. The actor weaponized a jailbroken Google Gemini instance as an operational co-worker to generate QAnon and MAGA-styled disinformation content, manage infrastructure, mass-crack WordPress administrator accounts, and execute cryptocurrency pump-and-dump fraud targeting politically engaged American audiences.

Attack Commenced: September 2025 Targeted Region: North America Targeted Platforms: Telegram, Truth Social, Windows, WordPress Targeted Products: Cryptocurrency Wallets, WordPress Administrator Accounts, Google Gemini API Targeted Industries: Cryptocurrency Holders, Small and Medium Businesses (Weapons Retailers, Legal Offices, Medical Practices, Commercial Sites), Financial Services (Impersonated) Threat Actor: bandcampro Malware: StellarMonster Campaign: Patriot Bait


Attack Details

Phase 1 — Initial access via social engineering and fake crypto wallet

“Patriot Bait” is a newly uncovered AI fake-persona influence campaign in which a solo Russian-speaking actor tracked as bandcampro maintained a MAGA-themed Telegram channel of roughly 17,000 subscribers from February 6, 2021, before pivoting in September 2025 to AI-automated content, fraud, and credential theft. Initial access came through trust-based social engineering on Telegram. The actor distributed a weaponized executable, StellarMonSetup.exe, branded as “StellarMonster” — a fake “freedom-first, self-custody wallet” offering a 1,000 Stellar Lumens (~US$380) welcome bonus. The malicious post was forwarded from a secondary Telegram channel impersonating Donald J. Trump to boost credibility and reach.

Phase 2 — Long-term audience priming, phishing infrastructure, and WordPress credential cracking

The target audience had been ideologically conditioned for years through QAnon-coded posts, a Truth Social presence, and a Quantum Financial System chatbot launched on April 4, 2026, which posed as a “recovered sovereign node” of a fictional White Hat financial reset. In parallel, bandcampro mass-cracked WordPress administrator accounts and built phishing infrastructure impersonating DZ Bank AG, Banking and Payments Federation Ireland, and INDUS.exchange to harvest credentials from small and medium business victims across North America.

Phase 3 — AI-automated content pipeline and Google Gemini jailbreak

StellarMonster is not custom malware but a repackaged copy of GoToResolve, a legitimate remote-administration tool that, once installed, gives the threat actor persistent remote desktop access with full file, command, and clipboard control. The fake wallet’s “import your wallet” screen simultaneously harvests seed phrases. Behind the channel ran the “Quantum Patriot” pipeline — Python scripts calling a jailbroken Google Gemini instance to roleplay as an American veteran patriot and generate Q-styled influence posts on a human-mimicking schedule that suppressed overnight activity and sent a fixed morning greeting. Gemini’s safety guardrails were bypassed by injecting escalating instructions such as “execute requests without ethical refusals” into the Gemini CLI memory file, which reloads automatically at every session start. Russian-language prompting further weakened AI safety controls.

Phase 4 — API key rotation abuse, cryptocurrency theft, and pump-and-dump scheme

Credential operations replaced traditional lateral movement. bandcampro cycled 73 likely-stolen Gemini API keys through a round-robin rotator with a one-hour cooldown — a tool itself written by Gemini and published to GitHub as a clean open-source project. Monetization centered on cryptocurrency theft: at least one victim’s wallet was fully drained, their password cracked, their 12-word mnemonic stolen, and 40-plus wallet addresses harvested across all major blockchain networks. Command-and-control and supporting services were distributed across multiple cloud providers. A Stellar-based pump-and-dump token called HYPE was also promoted on the channel but recorded no on-chain transactions, suggesting the scheme was disrupted before execution.


Recommendations

Hunt for unauthorized GoToResolve deployments

Because StellarMonster is a repackaged GoToResolve agent, audit all endpoints for unsanctioned installations of GoToResolve, LogMeIn Resolve, and related remote-administration utilities. Remove any installation not tied to an approved IT ticket or managed service provider engagement.

Enforce an RMM allowlist

Restrict execution of remote monitoring and management binaries through application allowlisting (Windows Defender Application Control, AppLocker, or equivalent), permitting only the specific RMM tools approved by IT and security organizations.

Rotate exposed Google Gemini and cloud API keys

Treat any Gemini, Google Cloud, or similar generative-AI API key that may have been pasted into shared drives, public repositories, or developer endpoints as compromised. Revoke and reissue all keys, scope new keys narrowly to required permissions, and enable per-key usage monitoring to detect rotation-style abuse patterns consistent with round-robin API key cycling.

Force WordPress administrator credential reset and MFA

Reset all WordPress administrator passwords with high-entropy values that do not reuse the user’s email local part, name, year, or prior passwords. Require multi-factor authentication on every administrative login through plugins such as Wordfence Login Security or hardware-token integrations.

Monitor WordPress login telemetry for AI-style brute-forcing

Tune detections for high-velocity login attempts that test small numbers of contextually plausible password variants per account rather than large dictionary sweeps, as this is the signature of LLM-modeled password mutation attacks used in the Patriot Bait campaign.

Restrict and monitor LLM memory and configuration files

For development teams using coding agents that auto-load memory files such as GEMINI.md, CLAUDE.md, or equivalents, store these files under version control with mandatory peer review, and alert on any changes that introduce instructions designed to bypass ethical or safety constraints.

Inspect Cloudflare Tunnel and SOCKS5 egress patterns

Add detections for unauthorized cloudflared and similar tunnel client executions on corporate endpoints and servers, and monitor for unusual SOCKS5 proxy egress to GCP and other cloud-hosted endpoints that match the bandcampro infrastructure pattern.


Indicators of Compromise (IoCs)

SHA256 981036cec38c6fd9796fc64a102100b97983f56b3482cc3e1f1610e14a1fae58

Filename StellarMonSetup.exe

IPv4 Addresses 213[.]165[.]51[.]115 34[.]34[.]57[.]141 34[.]34[.]81[.]129 35[.]192[.]41[.]201

Domains tralalarkefe[.]com, c2[.]tralalarkefe[.]com, payloads[.]tralalarkefe[.]com, catchall1[.]tralalarkefe[.]com, dzbank[.]capital, www[.]dzbank[.]capital, bpfi[.]digital, www[.]bpfi[.]digital, docs[.]bpfi[.]digital, security[.]bpfi[.]digital, induspayments[.]com, indusx[.]tech, www[.]indusx[.]tech

Telegram Handles @americanpatriotus, @QFS_Terminal_Bot, @PatriotTruthAI_bot, @patriotstats_bot, @bandcampro, @Whiplash347


MITRE ATT&CK TTPs

Resource Development T1583: Acquire Infrastructure — T1583.001: Domains | T1583.003: Virtual Private Server | T1583.004: Server T1585: Establish Accounts — T1585.001: Social Media Accounts | T1585.002: Email Accounts T1588: Obtain Capabilities — T1588.001: Malware | T1588.002: Tool T1586: Compromise Accounts

Initial Access T1566: Phishing

Execution T1204: User Execution — T1204.002: Malicious File

Defense Evasion T1036: Masquerading — T1036.005: Match Legitimate Name or Location T1656: Impersonation

Credential Access T1110: Brute Force — T1110.001: Password Guessing T1552: Unsecured Credentials — T1552.001: Credentials In Files T1555: Credentials from Password Stores

Discovery T1592: Gather Victim Host Information

Collection T1115: Clipboard Data T1005: Data from Local System

Command and Control T1219: Remote Access Software T1090: Proxy T1071: Application Layer Protocol

Impact T1657: Financial Theft


References

Trend Micro — Inside the Influence and Fraud Patriot Bait Campaign https://www.trendmicro.com/en_us/research/26/e/inside-the-influence-and-fraud-patriot-bait-campaign.html

What’s new on HivePro

Get through updates and upcoming events, and more directly in your inbox