Therac-25: History's Worst Software Error

The Machine

Radiation therapy has long been a cornerstone in the fight against cancer. By bombarding malignant cells with high-energy particles or photons, it disrupts their DNA and halts their spread. This treatment is essential for nearly 40% of cancer cases, often reducing or eliminating the need for invasive surgeries.

 In 1976, Atomic Energy of Canada Limited (AECL) made a significant advancement in this field with the development of a double-pass linear accelerator. This innovation allowed for more efficient treatment and resulted in a more compact machine. The Therac-25, released in 1983, was the pinnacle of this design—a fully software-controlled device that replaced many traditional mechanical safeguards with code. 

However, the software was dangerously underdeveloped. It had been adapted from the older Therac-20 model and was written by a single programmer. Eager to promote their new technology, AECL began selling the Therac-25 without thoroughly testing the software that controlled it.

The Therac-25 Tragedy: How a Software Glitch Changed Medical History

On a warm, clear June day in 1985, 61-year-old manicurist Katie Yarborough prepared for what should have been a routine cancer treatment. It was her twelfth session, and she drove herself to the Kennestone Regional Oncology Center in Marietta, Georgia. There, she would receive treatment from one of the most advanced machines of its time—the Therac-25, a linear accelerator designed to deliver targeted doses of radiation with precision and minimal discomfort.

The machine was intended to direct about 200 rads of radiation to her upper left chest in mere seconds to treat her lymph nodes without pain or complications. However, that day, something went terribly wrong. Instead of the expected absence of sensation, Yarborough felt a severe burning sensation. “You burned me,” she told the technician, who quickly dismissed her concerns, assuring her that the machine couldn’t malfunction in that way. 

In the weeks that followed, Yarborough's condition deteriorated alarmingly. The radiation had been so intense that it caused irreversible tissue damage. Her left breast, initially cancer-free, had to be removed, and the radiation destroyed the nerves in her left arm, leaving it permanently paralyzed. Despite these devastating injuries, Yarborough tried to maintain her independence, continuing to drive and live as normally as possible—until five years later, when she died in a car crash on a Georgia highway. Her name would later become synonymous with one of the worst software-related medical accidents in history.

Where did the Therac-25 go wrong?

At the heart of the Therac-25's failures were critical software bugs. Unlike its predecessors—the Therac-6 and Therac-20—the Therac-25 was designed to rely entirely on software for control, without the hardware safety mechanisms that earlier models included.

 Some of the software used in the Therac-25 was carried over from those earlier machines. While those bugs existed in earlier systems, they weren’t as dangerous because the hardware safety interlocks provided an extra layer of protection. In the Therac-25, however, the absence of these safeguards meant that the same bugs could have deadly consequences. 

One serious flaw allowed for a dangerous scenario: if the operator switched the beam setting from x-ray to electron beam within about eight seconds, the machine would still deliver the radiation dose but incorrectly display a message stating that no dose had been administered. This misleading message could prompt the operator to unknowingly deliver multiple high doses in succession, believing none had been given.

The Fallout

Katie Yarborough was the first confirmed victim of the Therac-25. Just two weeks after her injury, medical physicist Tim Still investigated her case. He found clear evidence of a massive radiation overdose: a red, dime-sized mark on her chest and a significant burn on her back, consistent with exposure to roughly 20,000 rads—100 times the intended dose. 

Still reported his findings to AECL, but the company dismissed the possibility of a malfunction, insisting that the Therac-25 couldn’t have delivered that much radiation. However, the evidence—and the emerging pattern—was undeniable. Within seven weeks, a second patient in Ontario, Canada, reported tingling and electrical shock during her treatment for cervical cancer. She, too, had been massively overdosed due to the software.

Other victims

Between 1985 and 1987, the Therac-25 was responsible for at least six severe radiation overdoses, resulting in the deaths of three patients. One of the victims, Ray Cox, received a dose that was 100 times greater than intended during treatment in Georgia and later died from complications related to the radiation. In Texas, Veronica Gildea suffered horrific burns from a massive overdose and died shortly after. Jeannine Gilbert, who was treated in Ontario, received a fatal dose to her shoulder and passed away months later after enduring significant pain and tissue damage.

The Aftermath

To make matters worse, an internal FDA memo revealed that AECL lacked formal software specifications and had no structured test plan for the Therac-25. Furthermore, the software was never reviewed by independent testers—an oversight that might have helped address internal biases and assumptions within the development team. Most of the software development and testing was carried out using a hardware simulator since testing directly on the real machine was challenging and potentially unsafe.