In July 2026, a joint investigation by the Associated Press and PBS Frontline offered the first systematic account of how AI is pushing telecom fraud out of the era of "low-tech, scattergun spam" and into one of "AI-driven industrial production." Its central finding points to a deeper, structural problem: the technologies and infrastructure of American tech companies — from AI models to satellite internet to ISP carriage — have become systematically embedded in the digital supply chain of the global fraud industry, while the United States' legal and regulatory framework has so far failed to mount an effective response.
The Industrialization of Fraud: How AI Reshapes the Operating Scale of Crime
The investigation's central subjects are Myanmar's scam compounds. These criminal factories in lawless territory are nothing new — but the arrival of AI has fundamentally redrawn their operating boundaries. Trafficked scammers working inside the compounds use software platforms called "Kongtian Smart Customer Acquisition" (KT) and "Global Social Traffic Navigator" (007TG), whose core engines are OpenAI's ChatGPT and Google's Gemini.
AI's role is not auxiliary but architectural. Real-time translation spans more than 100 languages; automated reply systems can manage dozens of fabricated identities at once; role-playing chatbots can read a target's social-media profile and automatically generate a tailored conversational strategy. One Indian man trafficked into Myanmar, Safeer Mohammed Koorimannil, working under coercion, locked onto roughly 50,000 targets across 17 countries in a single month. Set against the backdrop of one person simultaneously operating dozens of identities, that figure no longer describes "fraud" — it describes a production line.
Blockchain analysis supplies an even colder quantitative dimension: a single cryptocurrency wallet used by 007TG received $860,000 in payments between April 2024 and December 2025, and the fraud networks using these tools have profited by at least $75 million. Experts warn that AI is driving fraud toward full automation — once a large language model can independently carry out the entire workflow, from target selection and trust-building to steering the money transfer, the compounds may no longer need human operators at all.
The U.S. Federal Trade Commission estimates that Americans lost nearly $200 billion to fraud in 2024 — a figure approaching one percent of U.S. GDP that year, and larger than the defense budgets of most countries. What the investigation reveals is only the tip of the global fraud industry's iceberg.
The Digital Shadow of Physical Infrastructure: The Gray Role of Starlink and ISPs
Among the investigation's most unsettling findings is the pivotal role of American internet infrastructure in the global fraud industry. A sample analysis of device connections at four Myanmar scam compounds showed that one-fifth of the signals were carried by U.S.-registered companies, including Cogent Communications, AT&T, Oracle and DigitalOcean. These companies are not directly involved in the fraud, but they lack any legal or commercial incentive to proactively monitor and block abuse — "the cost of facilitating fraud is zero," as one cybersecurity expert summed it up.
Starlink's role is especially conspicuous. Although SpaceX has never officially sold service in Myanmar, Starlink has become the country's largest internet service provider, with a market share approaching 20 percent. The investigation found that at least 25 new scam compounds were built after the crackdown in autumn 2025, and 13 of them used Starlink to get online. In October 2025, Starlink severed service to more than 2,500 terminals near the compounds, and user numbers plunged — only to rebound a mere two months later. Analysts note that Starlink has the capability to cut service, but the fraud networks contribute heavily to its user base, leaving little commercial motivation to disconnect them.
This is not a situation that "technological neutrality" can simply wave away. When a private company's satellite internet becomes the channel of choice for the global fraud industry, and that company, fully aware of the fact, lacks the commercial incentive to shut it down, "neutrality" becomes a form of de facto complicity. The United Kingdom, the European Union, Australia and Singapore have all enacted regulations requiring tech companies to proactively prevent fraud, on pain of fines — the United States, to date, has no comparable framework.
American tech companies sit in a position of "able to stop it, but unwilling to." Cutting off fraud traffic dents revenue; identifying abuse demands an investment of manpower — and no law compels them to do either.
The Regulatory Vacuum and the Institutional Dilemma of American Tech
The contradiction the investigation exposes is, at root, structural: American tech companies sit in a position of "able to stop it, but unwilling to." Cutting off fraud traffic dents revenue; identifying abuse demands an investment of manpower; and no law compels them to do either. The "Scam Compound Strike Force" convened by U.S. Attorney Jeanine Pirro in Washington, D.C., working with Meta, SpaceX, Google and others, disrupted more than 1.4 million accounts and malicious traffic flows in May 2026 — but this reads more as a one-off enforcement performance than a systematic institutional response.
After the AP shared its findings, OpenAI banned three abusive accounts, while SpaceX and Starlink did not respond to detailed inquiries. The gulf between the scale of that response and the scale of the industry the investigation reveals is, in itself, a signal.
There is a contrast worth noting here: American tech companies are extraordinarily sensitive on the question of "transparency in cooperation with foreign governments," yet when their technologies are systematically put to use for fraud across the globe, the intensity of their intervention falls far short of public expectations. This is not a question of technical capability — Starlink can cut off 2,500 terminals, but it did so only when public pressure was at its peak, and the users were back within two months.
When Fraud No Longer Needs People
The most unsettling long-term signal in the investigation is this: AI is driving fraud toward "full automation." Once large language models can autonomously complete the entire fraud workflow, the compounds may no longer need trafficked human operators. That means the principal tool the international community now uses to fight fraud — severing the human-trafficking chain — may lose its efficacy in the future. If the scammers are all played by AI, the focus of enforcement will shift from "rescuing trafficked operators" to "preventing AI models from being abused" — and that requires tech companies to build anti-abuse safeguards into the product-design stage, rather than issuing bans after the fact.
The trouble is that American tech companies currently lack any institutional incentive to do so. Against a backdrop of regulatory vacuum, the judgment that "the cost of facilitating fraud is zero" applies not only to ISPs but equally to AI model providers. And as AI capabilities keep advancing, that zero-cost window is growing more dangerous by the day.