Investigation · OSINT · EDD · 2025

The Digital Arms Race: DarkGPT & Cheating‑as‑a‑Service for OSINT & EDD

How uncensored AI tools, leaked database engines, and invisible assistants are reshaping corporate intelligence, compliance, and the very meaning of digital trust.

By ·correspondent.cash·Toolkit··25 min read
$10.5T
Annual cybercrime cost (2025)
$19.9M
Lost per minute to cybercrime
200%
Increase in Dark AI mentions (2024)
$120M
Cluely valuation (June 2025)

In the prevailing corporate narrative, Artificial Intelligence is the ultimate productivity booster. But that same technology is being simultaneously weaponized as a digital super-weapon. We have entered an era where the world moves too fast for manual oversight—and sophisticated cybercrime is no longer the exclusive domain of elite state actors. It has been democratized, industrialized, and turned into a subscription service.

The contemporary intelligence landscape is undergoing a radical transformation driven by the proliferation of specialized, often uncensored, Large Language Models and automated reconnaissance frameworks. Tools such as DarkGPT and services under the “Cheating-as-a-Service” umbrella represent a significant shift in how both offensive actors and defensive researchers approach Open-Source Intelligence (OSINT) and Enhanced Due Diligence (EDD).

These instruments leverage the analytical power of models like GPT-4-200K to process vast quantities of unstructured data from leaked databases, social platforms, and corporate registries. What was once the sole domain of well-funded intelligence agencies is now accessible via subscription-based models or open-source GitHub repositories.

The same technologies that drive innovation can be weaponized for criminal purposes. The distinction between “White-Hat” and “Black-Hat” tools is increasingly narrow, defined largely by the user’s intent.

The global OSINT market is experiencing significant growth—size estimates range from $4.6 billion to $14.85 billion in 2024, with projections showing a CAGR of up to 28.2% through 2033. This market explosion is driven by cybersecurity needs, national security imperatives, and a new generation of investigators who cut their teeth on openly available data.

DarkGPT: The Bridge to Leaked Data

DarkGPT is not a newly trained model. It is a specialized, “jailbroken” variant of GPT-4-200K that acts as an analytical bridge between raw leaked data and actionable intelligence. It integrates directly with the Dehashed API to allow users to query massive databases of leaked credentials using plain English.

By simply asking, “What are the leaked passwords associated with this domain?”, a researcher—or an attacker—can instantly synthesize millions of rows of breached data into a targeted intelligence report. The tool automates repetitive OSINT tasks that would otherwise take analysts hours or days of manual database querying.

DarkGPT Technical Profile

Installation and Setup for Professionals

To deploy DarkGPT in a research or defensive environment, practitioners need Python 3.8+, Git, a Dehashed API key, and an OpenAI API key. The minimum recommended hardware is a quad-core processor with 16 GB RAM.

Terminal — DarkGPT Installation
# Clone the repository
git clone https://github.com/luijait/DarkGPT.git
cd DarkGPT

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env

# Edit .env and add your API keys:
# DEHASHED_API_KEY=your_key_here
# DEHASHED_USERNAME=your_email
# OPENAI_API_KEY=your_openai_key

# Launch
python3 main.py

The Underground Market for Malicious AI

DarkGPT exists within a broader ecosystem of “Dark AI” tools that have seen a 200% increase in market mentions in 2024. Operating under an AI-as-a-Service model, these uncensored systems sell subscriptions from $60 to $700 per month, making sophisticated cyberattacks affordable for novice criminals.

The threat actors don’t build models from scratch. Instead, they adapt powerful, often open-source models by removing ethical guardrails through prompt injection, fine-tuning on malicious data, or constructing wrapper applications that bypass safety filters. These jailbroken services are sold on Telegram channels and dark web forums, sometimes for as little as €60 per month.

The world loses approximately $19.9 million per minute to cybercrime. This translates to roughly $1.2 billion every hour. The clock is no longer just a measure of time—it is a measure of loss.

The Arsenal: Dark AI & CaaS Services Compared

A direct comparison of primary services, their accessibility, pricing, and operational intent within the OSINT and EDD domains.

ServicePrimary FunctionArchitecturePricingIntentKey Features
Toolkit15+ Free Tools ListedCuratedClick&DiscoverDehashed API; CLI; credential analysis
DarkGPTLeaked database querying via natural languageJailbroken GPT-4-200KOpen-source + APIDefensive OSINTDehashed API; CLI; credential analysis
CluelyReal-time invisible AI assistantLLM extension (Claude/GPT-4o)Free / $20/moAmbiguous CaaSUndetectable; screen+audio; real-time prompts
WormGPTPhishing & BECJailbroken GPT-J$60–$700/moMaliciousPersonalized phishing; executive mimicry
FraudGPTMalware & hacking toolkitJailbroken LLM$200/mo–$1.7K/yrMaliciousExploits; vulnerability research; carding
EvilAIMalware & phishing pagesGPT-4 + Mistral$60–$700/moMaliciousTrained on malware code; stealth scripts
XanthoroxAutonomous cybercrimeCustom 5-model system~$200–$300/moMaliciousCoder/Vision/Reasoner; voice clone; darknet
DarkBERTDark web classificationBERT on Dark Web dataResearchDefensiveThreat keywords; illicit community monitoring

Cheating-as-a-Service: The Erosion of Digital Trust

A major development in the 2025 landscape is “Cheating-as-a-Service” (CaaS)—a grey market of tools that blur the lines between human augmentation and unethical advantage. The flagship product is Cluely, an “invisible AI assistant” designed to be completely undetectable during meetings, calls, and exams.

Cluely analyzes both on-screen visual data and audio in real-time, providing answers during coding interviews, high-stakes exams, and critical sales calls. Its founders, Roy Lee and Neel Shanmugam (both 21), were expelled from Columbia University for creating the tool’s predecessor. Despite this notoriety, Cluely raised $15 million from Andreessen Horowitz in June 2025, reaching a $120 million valuation.

Cluely at a Glance

The tool’s explosive success points to a broader cultural shift that researchers describe as “financial nihilism”—where efficiency and outcome are prioritized over traditional ethics. In an EDD context, this poses a significant risk: corporate executives and technical candidates can now project competencies they don’t actually possess.

The Counter-Market Emerges

Cluely’s rise has triggered an immune response. Startups like Validia (“Truely”) and Proctara are developing detection tools for AI-assisted cheating. Universities are shifting from prohibition to managed integration, with 89% of students reportedly using AI tools by 2025.

The cycle: provocative AI product → product-market fit → top-tier VC funding → defensive counter-market → institutional policy adaptation. This pattern will repeat across industries.

The Ferrari Deepfake Case

In July 2024, a Ferrari executive narrowly escaped a deepfake scam where an attacker used AI voice cloning to mimic CEO Benedetto Vigna’s Southern Italian accent in a WhatsApp call, requesting urgent funds for a “confidential acquisition.” The executive only detected the fraud by asking a personal question the AI couldn’t answer.

Watch & Listen

The full deep dive in video, audio, and slide formats.

▶ YouTube — Video Deep Dive
🎧 Spotify — Follow The Money Podcast
📊 SlideXchange — Presentation Slides

If the slides don’t load, open directly on SlideXchange →

AI-Enhanced Attack Vectors vs. Defensive Responses

The application of DarkGPT and CaaS tools in Enhanced Due Diligence responds to the “wobbly façade” of traditional banking compliance. Organizations now face threats like autonomous ransomware that analyzes exfiltrated financial data to calculate ransom based on the victim’s liquidity.

Offensive

Hyper-Personalized Phishing

AI reduces personalized phishing costs by 95%. Criminals mimic executive writing styles and deploy voice cloning for deepfake calls.

Defensive

Synthetic Text Detection

Training detection models against AI-generated text, identifying patterns that distinguish machine-crafted from genuine messages.

Offensive

Synthetic Identities & Deepfakes

Generating synthetic IDs and deepfakes to bypass biometric checks and KYC. Tools like DeepFaceLab increasingly fool verification.

Defensive

Red Team Toolkit

Open-source tools simulating Dark Web AI attacks for defensive testing. Standardized threat schemas and deepfake classifier benchmarks.

Offensive

Automated Reconnaissance

DarkGPT automates credential stuffing and target network mapping, moving from discovery to exploitation at machine speed.

Defensive

AI-Powered SOC Monitoring

LLM agents for autonomous threat response—reducing response from 3–7 hours (manual) to 1–2 minutes (AI-driven).

The “Agentic AI” Threat

Agentic AI” systems autonomously execute multi-stage attacks. They perform reconnaissance, harvest credentials, penetrate networks, and generate psychologically targeted ransom notes from exfiltrated data—all without human intervention.

Case Study: Jan Marsalek & the Wirecard Investigation

Investigators Christo Grozev and Roman Dobrokhotov used OSINT “follow the money” approaches to uncover fugitive Wirecard COO Jan Marsalek’s ties to Russian intelligence. This case demonstrates the profound transformation of investigative journalism through digital sleuthing. Listen on Spotify.

Strategic Defense Framework for 2025

Blockchain for Data Provenance

Immutable ledger to track data origins and prevent “data poisoning” during AI model training.

Federated Learning

Distributed model training without centralizing raw data, reducing breach “blast radius.”

Zero-Trust Architecture (ZTA)

“Never trust, always verify.” Continuous authentication with AI-monitored behavioral analysis.

First-Order Logic (FOL) for Compliance

Deterministic, machine-checkable proofs for mathematically verifiable compliance without “black-box” AI.

Self-Sovereign Identity (SSI)

Verified credentials in private digital wallets. “Verify once, use everywhere” via identity.global and Nansen.ID.

The Professional Investigator’s Toolkit

35+ essential services for OSINT, EDD, and defensive intelligence—each with direct links to get started.

OSINT · Leaked Data

DarkGPT

GPT-4-powered OSINT assistant for querying leaked databases via natural language. Dehashed API integration.

github.com/luijait/DarkGPT
OSINT · Breach Data

Dehashed

Primary breach data search engine powering DarkGPT. Search leaked credentials for security auditing.

dehashed.com
OSINT · Email Intel

H8Mail

Scans emails against databases for leaked passwords. Essential for credential reuse risk assessment.

github.com/khast3x/h8mail
OSINT · Email Footprint

Holehe

Checks email presence on 120+ platforms. Maps digital footprint from a single email address.

github.com/megadose/holehe
Identity · Username

Maigret

Username searches across 3,000+ sites. The go-to tool for username OSINT reconnaissance.

github.com/soxoj/maigret
Identity · Google Intel

GHunt

Intelligence from Gmail: owner name, Google service activity, calendar-based location data.

github.com/mxrch/GHunt
Identity · Dorking

Mr. Holmes

Google Dorks for phone numbers, domains, IP addresses. Local database for organized investigations.

github.com/Lucksi/Mr.Holmes
Identity · Face Search

PimEyes

Reverse image face search. Find where a face appears across the internet.

pimeyes.com
Identity · Face Search

Search4Faces

Face search across VK and OK social networks. Popular for CIS region investigations.

search4faces.com
Social · Link Analysis

Maltego

Data mining and link analysis. Visualizes relationships between people, groups, websites, and profiles.

maltego.com
Social · Automated Recon

SpiderFoot

Automated collection from 200+ sources for reconnaissance and footprint mapping.

github.com/smicallef/spiderfoot
Social · Harvesting

theHarvester

Gathers emails, subdomains, hosts, employee names from public sources.

github.com/laramies/theHarvester
Social · Multi-Platform

Social Links

Aggregates 500+ sources: social media, messengers, blockchains, Dark Web.

sociallinks.io
Dark Web · Threat Intel

Flashpoint

Actionable intelligence from illicit communities. CTI, physical security, brand protection.

flashpoint.io
Dark Web · Search

DarkOwl

Automated dark web marketplace, forum, and encrypted chat search.

darkowl.com
Dark Web · Breach Intel

Constella Intelligence

World’s largest breach database (1T+ attributes). Deep and Dark Web identity intelligence.

constella.ai
Cyber · Threat Intel

Recorded Future

Enterprise AI/ML threat intelligence from open sources, dark web, and technical feeds.

recordedfuture.com
Cyber · Law Enforcement

ShadowDragon

OSINT for law enforcement. Dark web forums, social media, encrypted messaging.

shadowdragon.io
Cyber · Forensics

Cellebrite

Industry-standard forensic data extraction from physical devices.

cellebrite.com
Infrastructure · IoT

Shodan

Search engine for internet-connected devices, exposed servers, IoT, and SCADA.

shodan.io
Infrastructure · Surface

Censys

Global attack surface index for internet-connected devices and vulnerability discovery.

censys.io
Infrastructure · Asia

ZoomEye

Chinese search engine for internet-connected devices. Asian network infrastructure research.

zoomeye.org
Blockchain · AML

Chainalysis

Pioneer in blockchain analysis for tracking illicit crypto flows and AML compliance.

chainalysis.com
Financial · Due Diligence

Sayari Labs

Corporate registries, legal filings, sanctions lists for risk mapping and ownership analysis.

sayari.com
Financial · Sanctions

ComplyAdvantage

AI-driven AML, sanctions screening, and adverse media monitoring for KYC/EDD.

complyadvantage.com
Geospatial · Satellite

Maxar Technologies

Ultra-high-resolution (30 cm) satellite imagery for defense and investigative journalism.

maxar.com
Geospatial · Real-Time

Sentinel Hub

Real-time satellite imagery access. Democratized GEOINT without expensive licensing.

sentinel-hub.com
Geospatial · Flights

Flightradar24

Real-time flight tracking for private jet investigations and movement analysis.

flightradar24.com
OSINT · Metadata

FOCA

Hidden metadata extraction from Word, PDF, PPT documents.

github.com/ElevenPaths/FOCA
OSINT · Historical

Wayback Machine

Billions of archived web pages. Find deleted content and corporate history.

web.archive.org
OSINT · Verification

InVID / WeVerify

Video/image verification. Detects manipulated media and combats deepfakes.

invid-project.eu
Privacy · Removal

DeleteMe

Personal data removal from data broker sites.

joindeleteme.com
Privacy · Removal

Incogni

Automated data removal from brokers by Surfshark.

incogni.com
Narrative · AI Detection

Blackbird.AI

Detects harmful narratives: brand risk, stock manipulation, geopolitical threats.

blackbird.ai
Framework · Directory

OSINT Framework

Community-maintained directory of OSINT tools. The starting point for any investigation.

osintframework.com
Framework · Curated

Awesome OSINT

GitHub curated list of OSINT resources, tools, and techniques.

github.com/jivoi/awesome-osint
CaaS · Invisible AI

Cluely

The defining Cheating-as-a-Service product. Undetectable real-time AI assistant.

cluely.com
Detection · Anti-Cheat

Validia (Truely)

Detects AI-assisted cheating during video interviews. Defensive counter to Cluely.

validia.ai

The “ShmagunGPT” Framework

The “ShmagunGPT” concept digitizes the investigative expertise of top journalists like Olesya Shmagun—“understanding the origin of money” and “spotting anomalies in behavior.” This methodology, from Navalny’s anti-corruption team, was built on open data. Read more: From Dissident to Detective.

The most important qualities for modern OSINT aren’t a special degree or clearance level. They are patience and attentiveness—because the data is already out there, waiting to be found.

Conclusion: Navigating the 2025 Intelligence Landscape

The dual-use nature of specialized AI instruments represents the defining challenge for corporate security in 2025. The future of EDD lies in self-sovereign identity and deterministic compliance based on First-Order Logic. Trust must be cryptographically proven and constantly re-verified against AI-powered deception.

All Source Links & References