This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
An investigation by C4ADS and the Associated Press revealed that criminal scam compounds in Southeast Asia are using AI-powered software suites, such as 007TG and KT Smart Translation, to industrialize fraud. These tools integrate leading LLMs like ChatGPT and Gemini to automate multi-language messaging, roleplay characters, and monitor trafficked workers. The platforms have facilitated tens of millions of dollars in illicit proceeds, while the scam compounds themselves rely on a global internet infrastructure that includes U.S.-based ISPs and Starlink.
007TG was a software suite reportedly used by scam compounds to run fraud across social media, using AI-assisted messaging and translation to help operators manage targets while concealing their identities. Wallet records linked the platform to at least four known scam networks that collectively received at least $75 million, though the evidence does not show how much was directly attributable to 007TG.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
Also classified as
Victims anthropomorphized and placed trust in the AI-generated personas, leading to severe emotional and material dependence.
The primary aspect of the incident is the coordinated use of AI systems to conduct targeted romance scams and financial fraud for personal and organizational gain.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The harm is caused by human malicious actors intentionally abusing deployed AI models to conduct large-scale financial fraud and monitor trafficked workers.
Added categories peak at 3: Substantial (Psychological Harm, Inferred)
Criminal syndicates in Southeast Asia are leveraging commercial LLMs to scale global financial fraud and monitor trafficked workers. While not a direct state-level threat, the integration of AI into transnational organized crime represents a substantial escalation in economic harm and human rights abuses.
The AI systems (ChatGPT, Gemini) function as assistants that augment the scammers' capabilities by translating messages, generating automated replies, and roleplaying characters, requiring human oversight and action to execute the scams.
Unacceptable Risk: The report describes 'AI used for subliminal manipulation or behavioral influence leading to harm' and 'exploitative systems targeting vulnerabilities' where scammers use AI to 'make each victim fall in love' and 'deceive targets' through 'manipulative language'.
The AI systems are primarily used as chatbots to generate automated replies and roleplay characters to deceive victims, and as translation tools to allow scammers to operate across over 100 languages.
The reports from AP and C4ADS provide extensive, highly detailed, and well-documented evidence of the AI models used, the specific software suites (KT and 007TG), the mechanisms of fraud, and the severe physical and financial harms caused. The findings are backed by blockchain analysis, device data, and first-hand interviews with victims and former scammers.
None. The role of AI in scaling and facilitating these scams is clear and undisputed in the reports.
Technology
Abuse of LLMs for Automated Replies
Scammers use ChatGPT and Gemini to automate deceptive messaging at scale.
AI Roleplay Chatbots for Deception
GPT-4 powered chatbots help scammers build highly convincing personas.
AI-Powered Employee Monitoring
007TG uses AI sentiment analysis to track victim emotional shifts and quotas.
Data Inputs
Difficulty Distinguishing Scams
Scam prompts mimic legitimate uses like translation and basic writing.
Evasion via Laundered IP Addresses
Scammers route traffic through US cloud services to hide physical locations.
Lack of Real-Time Prompt Flags
Input data lacks real-time flags to detect manipulative romance scam scripts.
Human Factors
Coerced Labor in Scam Compounds
Trafficked workers are physically forced to operate AI-powered scam systems.
Target Vulnerability to AI Personas
Targets are deceived by highly realistic AI-generated personas and text.
Cognitive Enablement of Scammers
AI tools enable low-skill scammers to assume highly technical roles.
Process and Methods
Reactive Abuse Mitigation Policies
AI companies and ISPs mostly take action after abuse reports are received.
Opaque Gray Market Software Sales
Software like KT and 007TG is sold via Telegram and paid with crypto.
Inadequate Proactive Monitoring
Lack of active systems to monitor and block API misuse in real-time.
Management
Prioritizing Growth Over Safety
Tech companies lack business incentives to invest heavily in scam prevention.
Weak Enforcement of Terms of Service
Failure to vigorously enforce terms of service banning fraudulent activities.
Inadequate API Access Controls
Management failed to restrict API access for gray-market scam software.
Regulatory Environment
Lack of Legal and Business Incentives
US tech companies face no financial cost for facilitating cyber scams.
Voluntary Cooperation Model in US
US regulators rely on voluntary cooperation rather than strict penalties.
No Utility-Level Safety Mandates
Tech and telecom providers are not legally responsible for content safety.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
