Anthropic Model Finds a Vulnerability in MyChart

MyChart is an electronic health records system that holds data on more than 320 million patients. It recently had a security problem. Epic Systems, the company that builds MyChart, paused new feature development for about six weeks to fix vulnerabilities in the system. That alone is worth noting, but what caught my attention is how the vulnerabilities were found. According to The New York Times, Epic's security chief, Sterling Martin, said the vulnerabilities were discovered using Anthropic's Mythos model. In some client configurations of MyChart, these vulnerabilities let outside parties access patient data without leaving a single log entry. Epic's CEO, Judy Faulkner, said the company now plans to make improving its product's security a priority.

What makes this story interesting is that the vulnerabilities were not found by a traditional pentesting team, but by an advanced AI model. I work with Claude Code every day, and I can confirm these models are genuinely strong at methodically working through different configurations, not just writing code. If you are working on a project that handles other people's data, consider running a similar check yourself, before someone else finds your vulnerability first.