OpenAI Shuts Down ChatGPT Accounts Powering a Cambodia-Based Scam Factory

OpenAI has shut down a coordinated network of ChatGPT accounts that powered a Cambodia-based scam factory running multi-vector fraud and trafficking-linked operations, and has shared indicators with industry peers and authorities to make the network’s reconstitution significantly harder.

Earlier in 2026, OpenAI’s threat intelligence team disrupted a scam syndicate that weaponized ChatGPT to drive investment, romance, gambling and law-enforcement impersonation schemes, following an initial lead from WhatsApp.

The operation was traced to a coordinated set of accounts very likely operating in or around Poipet, Banteay Meanchey province, a border region repeatedly linked in public reporting to online scam compounds and human trafficking operations.

OpenAI revoked the ChatGPT accounts involved, pushed threat signals to industry partners and relevant authorities, and implemented additional safeguards to prevent the network from regaining access to its AI services.

Operators in the Cambodia-based network used ChatGPT to spin up fake dating profiles and “investment expert” identities, translate and optimize scripts for WhatsApp and Telegram outreach.

Craft social media content that supported long-running relationships with targets before introducing bogus cryptocurrency and spot gold trades.

OpenAI’s investigation highlights a key reality of modern scam ecosystems: organized criminal groups no longer silo their activity into a single fraud vertical, but fluidly blend multiple narratives and personas to maximize victim conversion.

Other personas posed as representatives of online gambling platforms pushing fake bonuses, or as law enforcement officials coercing victims into paying fabricated fines for alleged serious offenses, often backed by AI-generated images of forged passports, legal notices and trading interfaces.

OpenAI Shuts Down ChatGPT Accounts

Consistent with patterns documented in OpenAI’s previous “Disrupting Malicious Uses of AI” reports from June and October 2025, the network followed a repeatable “ping, zing, sting” lifecycle in victim interactions.

The ping phase involved AI-assisted outreach, with ChatGPT generating and translating initial contact messages and helping operators research plausible dating and professional personas.

The zing phase relied on emotional engineering: promises of guaranteed, “risk-free” returns, intense romantic language, secrecy instructions, and manufactured urgency around expiring bonuses were all used to deepen trust and pressure victims.

Finally, the sting phase pushed victims to deposit funds for activation fees, unlock supposed rewards or settle fictitious legal penalties, then demanded screenshots and account details as proof, enabling rapid monetization across a high volume of targets.

Beyond pure fraud, OpenAI observed activity strongly suggestive of human trafficking and forced labor within the scam factory’s internal operations.

ChatGPT was used to generate social media advertisements for “chatter” jobs in Poipet that promised flights, accommodation, meals, visas and work permits language aligned with wider public reporting on Southeast Asian scam compounds that recruit on the pretense of legitimate employment before trapping workers in debt bondage and coercion.

Investigators also saw content documenting employee debts, salary deductions, disciplinary fines and loan repayments, alongside translations of discussions around visa overstays, work permits, escape attempts and potential criminal liability for trafficked workers, echoing findings from Amnesty International and U.S. Trafficking in Persons reports on Cambodian cyber-scam centers.

While precise loss figures remain unknown, internal communications referenced hundreds of targets and individual victims losing thousands of dollars across the blended scam types.

OpenAI has responded by banning all identified ChatGPT accounts linked to the operation, tuning its abuse-detection pipelines to the specific tactics and linguistic signatures observed, and sharing indicators and analytical insights with messaging platforms.

The case underscores a broader trend documented in OpenAI’s threat intelligence publications: large language models are increasingly integrated into scam and trafficking infrastructure, and effective mitigation now depends on continuous cross-industry intelligence sharing and proactive model-level safeguards.

$1M Data Breach Warranty is Genuine Protection?: Download 10 Point Free AI SOC Breach Warranty Guide