ChatGPT used to conduct crypto scams targeting US real estate agents at Myanmar-Thai border fraud compound
Kenyan jobseeker Duncan Okindo was trafficked to a fraud compound on the Myanmar-Thai border where he used ChatGPT to generate deceptive messages targeting US real estate agents with cryptocurrency scams. Okindo spent four months in the compound conducting "pig-butchering" fraud before being freed following Thai crackdowns.
What happened
Duncan Okindo, a 26-year-old Kenyan, was trafficked to Myanmar under false promises of a customer service job in Thailand. Upon arrival, he was abducted at the airport and taken to KK Park, a scam compound on the Myanmar-Thai border guarded by heavily armed men.
In the compound, Okindo worked in a large room with hundreds of other forced laborers using desktop computers. He used the free version of ChatGPT to craft messages targeting US real estate agents with fake cryptocurrency investment schemes, a scam type known as 'pig-butchering' where scammers cultivate victims' trust before stealing their money.
Okindo spent four months in the compound before being freed following Thai crackdowns. An anti-trafficking group, HAART Kenya, confirmed his rescue along with several other Kenyans from the facility.
Reported impact
US real estate agents were targeted with fraudulent cryptocurrency investment schemes. Okindo identified potential victims by trawling property websites including Zillow.com where agents advertise their services.
Okindo endured forced labor and abuse by gang bosses during his four months in the compound. He remains traumatized from the experience.
Index analysis
ChatGPT's free version was used at scale to generate deceptive messages in a coordinated fraud operation. OpenAI stated it 'actively works to identify and disrupt scam-related misuse of ChatGPT' and that the model refuses requests breaking its anti-fraud rules.
Reuters interviewed about a dozen forced laborers whose accounts conformed with Okindo's story, though Reuters could not independently verify all details of his account.
Governance controls
OpenAI stated that the AI's underlying model refuses requests that break its anti-fraud rules.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.