In the past, executing health insurance fraud was a complex task that required a deep understanding of medical terminology and billing codes. Fraudsters had to possess significant skill if they wanted to fabricate health records convincingly or deceive call center employees, necessitating phone calls where they impersonated patients or medical professionals.
However, advancements in artificial intelligence have drastically simplified these illicit activities. With just a prompt in a sophisticated language model like ChatGPT, individuals can now easily generate fake documentation for non-existent medical procedures. Furthermore, AI-driven systems can make thousands of automated calls to insurance companies each day without any human intervention.
Kurt Spear, vice president of financial investigation and provider review at Highmark, expressed concerns about how AI is increasingly being exploited for fraudulent purposes within the insurance industry. "We knew AI would be leveraged against us for fraud, and we are beginning to witness that reality," he noted.
The swift rise of AI poses substantial risks not only to private health insurers but also to Medicare, Medicaid, and other government-run insurance programs, as highlighted in a report projected for 2025. The National Health Care Anti-Fraud Association estimates that healthcare fraud costs the industry nearly $480 billion annually. Recovering these losses typically hinges on criminal investigations, which often yield only a fraction of the stolen funds.
While some organizations, like UPMC, reported minimal instances of AI-generated fraud attempts, other agencies, such as the Pennsylvania Department of Human Services, have encountered sporadic fake calls without any notable increase in fraudulent activities.
According to the National Health Care Anti-Fraud Association, AI tools can fabricate medical records, create false patient identities, impersonate healthcare providers, and exploit insurance policies for financial gain. The sheer scale of these emerging technologies is particularly alarming to both the insurance and cybersecurity sectors.
"Some of our clients have documented up to 15,000 bot calls over just a couple of months," stated Jason Barr, vice president of healthcare at Pindrop, a firm that provides technology utilized by leading health insurers to discern between human and AI-generated voices. Pindrop's AI detection software operates invisibly during calls, evaluating voice elements, behavior patterns, and other characteristics to identify the speaker's authenticity. It also incorporates data derived from signal carriers and the devices initiating the calls.
Barr noted a significant evolution in synthetic voice technology over the past two years. While many AI-generated voices were easily recognizable previously, they have since become much more sophisticated. Though not yet flawless—customers have reported instances of robocallers altering their accents or mimicking the agent's voice during calls—the improvement in realism is notable.
Highmark employs various mechanisms to detect AI-driven fraud and is currently integrating a new tool aimed at spotting anomalies in medical imaging at a granular level, suggesting that traditional human analysis may soon be insufficient in many cases. A recent study published in the journal Radiology indicated that radiologists could only distinguish between authentic and deepfake X-rays with a 75% accuracy rate.
On another front, researchers at the University at Buffalo are pioneering techniques to identify AI-generated radiology reports, which are the documented outcomes of imaging tests. Their findings reveal a key difference: large language models often use intricate and formal language, whereas doctors typically favor more straightforward communication.
Despite the introduction of advanced technologies to combat fraud, everyday patients remain one of the most effective lines of defense. If individuals receive notifications regarding medical services they did not receive, it is often an indication that something suspicious has occurred. Spear emphasized, “Some of the best referrals come from members.”
