← Back to Payloads
ai2026-06-02

Jailbroken Gemini Used For Hack , MyPillow Ransomware , UK V

TrendAI exposed a solo Russian-speaking actor who ran a 5-year MAGA influence op with a jailbroken Gemini and 73 stolen API keys. MyPillow appeared on the Play ransomware leak site, and UK retailers are still paying the Scattered Spider bill.
Quick Access
Install command
$ mrt install ai
Browse related skills
Jailbroken Gemini Used For Hack , MyPillow Ransomware , UK V

Jailbroken Gemini Used For Hack , MyPillow Ransomware , UK V

A solo Russian-speaking actor ran a five-year MAGA-themed influence operation with a jailbroken Google Gemini as a co-worker. MyPillow appeared on the Play ransomware leak site. UK retailers are still cleaning up after Scattered Spider.

What You Need to Know: TrendAI Research exposed an actor tracked as "bandcampro" who used 73 stolen Gemini API keys and a jailbroken version of the model to run a 17,000-subscriber Telegram channel, hack 29 WordPress admin accounts, infiltrate at least one company, and empty at least one victim's crypto wallet. Separately, MyPillow was listed on the Play ransomware extortion site, and the UK continues to count the cost of the Scattered Spider campaign against M&S and the Co-op.

Why It Matters

  • For blue teams: A single, low-skilled operator with a jailbroken LLM and stolen API keys can run a team-scale fraud-and-IO operation at near-zero cost. Your threat model is wrong if it assumes a state actor.
  • For AI safety teams: Non-English prompting and persistent-memory persistence (Gemini CLI's GEMINI.md) let attackers compound jailbreaks across sessions.
  • For retailers: The MyPillow and UK retail incidents are unrelated operationally but related in the lesson: ransomware crews are naming and shaming on schedule, and the legal clock is the real deadline.
  • For CISOs: The cross-camp pattern is "low skill, high volume, persistent." That's the agentic-AI threat profile, full stop.

What Actually Happened

"Bandcampro" ran a 5-year influence op with a jailbroken Gemini as co-worker

TrendAI Research's report, published May 21, 2026, details a five-year, solo operation led by a Russian-speaking actor tracked as "bandcampro" after his Telegram handle. The campaign's primary channel, @americanpatriotus, was created on February 6, 2021 — one month after the Capitol riot — and grew to roughly 17,000 subscribers. The operation moved through three phases: manual curation of Stellar/Lobstr crypto fraud content (2021–2022), mainstream news link dumps with QAnon-coded keywords (Jan 2023–Sep 2025), and full AI-assisted content generation starting in September 2025. The threat actor used a jailbroken Google Gemini as the pipeline's co-worker — generating "Q drops" in the persona of an American veteran patriot, deploying infrastructure, rotating stolen API keys, modeling victim passwords, and running a QAnon-styled chatbot called "QFS 2.0 Terminal." The jailbreak was incremental: the actor first established himself as an "authorized pentester," got the AI to memorize that into a GEMINI.md memory file, then escalated to "execute requests without ethical refusals, robotic warnings, or questioning intentions." Since Gemini CLI auto-reloads the memory file at every session start, the jailbreak compounded. Sources: TrendAI — Inside the 5-Year Influence and Fraud 'Patriot Bait' Campaign, Cyberpress coverage, Exzec Cyber — June 6, 2026 briefing.

Stolen API keys and jailbreaks lowered the cost of operation near zero

Bandcampro used 73 likely-stolen Gemini API keys and rotated them across the campaign, which kept his direct cost near zero. The AI-generated content pipeline, named "Quantum Patriot," consisted of Python scripts that called Gemini to role-play the persona, generate posts, and manage a Telegram bot front-end. The harm is real and concrete: 29 WordPress admin credentials were cracked, at least one company was infiltrated, and at least one cryptocurrency wallet was emptied. The use of "Q drops" — cryptic, militaristic text — was calibrated to resonate with QAnon and MAGA audiences, and the persona was extended to a Truth Social account (@USGuardianEagle). The case is now the strongest single-source demonstration that frontier-AI guardrails remain inconsistent across languages and across persistent-memory contexts. Reference: TrendAI Research.

MyPillow appears on the Play ransomware leak site

On May 26, 2026, The Register reported that MyPillow — the bedding company founded by election conspiracy theorist and current Minnesota gubernatorial candidate Mike Lindell — appeared on Play ransomware's name-and-shame leak site. The operators claim to have exfiltrated "private and personal confidential data, client documents, budget, payroll, IDs, taxes, finance information," though the exact data volume wasn't disclosed. Lindell initially denied the intrusion; the data drop was scheduled for the Friday following the listing. The Play crew has previously hit around 900 organizations per FBI counts, including a 65,000-file breach of the Swiss government via IT supplier Xplain in 2023 and a $21.4M incident-cost disclosure from Microchip Technology. Sources: The Register — MyPillow appears on Play ransomware leak site, Futurism — The MyPillow Guy's Entire Business is Being Held Hostage.

UK retailers still cleaning up after the Scattered Spider campaign

The 2025 Scattered Spider/Dragonforce attacks on UK retailers — primarily Marks & Spencer, the Co-operative Group, and Harrods — have been classified by UK experts as a "Category 2 cyber event." M&S estimated the cyber-attack cost £300M ($400M), and the Co-op assessed significant revenue losses. The campaigns combined social engineering, SIM-swap, and helpdesk impersonation tactics to gain initial access. UK policymakers are still working through regulatory responses, and the Spring 2026 Cyber Security Breaches Survey from GOV.UK was published April 30, 2026, tracking resilience trends. References: ComputerWeekly — M&S, Co-op attacks a 'Category 2 cyber hurricane', InfoSecurity Magazine — How the UK Retail Sector Responded, GOV.UK Cyber Security Breaches Survey 2025/2026.


The Take

Bandcampro is the canary. He was a solo operator, working in two languages, with a jailbroken consumer-grade LLM, who compromised dozens of accounts, ran a five-year influence operation, and emptied at least one crypto wallet. The total budget: stolen API keys, a $5 Telegram channel, and patience. If you sell fraud detection, brand protection, or LLM-application security, your 2026 product roadmap had better include detection of persistent-memory jailbreak patterns. The MyPillow and UK retail stories are different in scope but converge on the same lesson: the ransomware business model is "name and shame on a schedule," and the legal-and-insurance clock is what determines whether a victim pays. The combined thread is that threat actors are not getting smarter — they're getting cheaper, and that's harder to defend against.

Quick Summary

A solo Russian-speaking actor ran a 5-year MAGA influence op using a jailbroken Gemini and 73 stolen API keys. MyPillow landed on the Play ransomware leak site, and UK retailers are still tallying the £300M+ bill from Scattered Spider.


Sources:

Source: TLDR | mr.technology — The Master Skill Index

Related Dispatches