The biggest epidemic you’ve probably never heard of

On September 13, 2005, a software bug in the online role-playing game World of Warcraft unleashed something extraordinary: one of the most lifelike simulated disease outbreaks ever observed. Entirely by accident.

The episode began with the release of a new high-level game area, Zul’Gurub. At its center was a powerful boss named Hakkar the Souflayer. During battles, Hakkar cast a spell called Corrupted Blood, designed to drain players’ health quickly. The mechanic had an additional twist: the effect could spread from one character to another if they stood too close.

Inside the raid dungeon, where Hakkar lived, the spell worked exactly as intended. But the designers overlooked a crucial detail, animal companions.

In World of Warcraft, players can summon pets that fight alongside them, but the game also penalises their loss. Therefore, players can temporarily dismiss their pets, storing them in a kind of suspended state before summoning them again later.

This is exactly what player zero did. It dismissed its infected companion to resummon it later outside the dungeon, in a crowded city, and here’s the overlooked variable. The animal companion came back in the game carrying the disease.

Within hours the virtual pathogen escaped into the broader game world in the crowded cities where players of any level gathered. Soon after, lower level players started collapsing on the ground leaving streets filled with white skeletons. Soon after, the global chat was flooded with alarm as players tried to make sense of what was happening.

Among the millions of players roaming World of Warcraft, there were also a few epidemiologists. One of them, Eric Lofgren, quickly realised that the chaotic digital outbreak resembled real epidemics in ways that were difficult to ignore.

For epidemiologists, human behavior is the hardest variable to capture. Mathematical models are good at describing how pathogens spread under controlled assumptions. But real outbreaks are shaped by panic, altruism, denial and defiance. Factors that equations struggle to predict. And ethically, researchers cannot stage a real epidemic simply to observe how people react.

Yet, the Corrupted Blood incident effectively created one. At the time, World of Warcraft hosted millions of players. Real people making spontaneous decisions under pressure inside a shared virtual world.

The parallels with real outbreaks were striking.

Infected characters frequently traveled from remote regions to crowded cities before dying in search of help. Uninfected ones, moved to remote areas and instituted quarantines to avoid infection. The ones with healing powers rushed toward cities only to revive players that were re-infected (there was no immunity) and to get infected themselves. Even a handful deliberately spread the infection, summoning infected pets in crowded areas.

Panic, altruism, misinformation, rule-breaking, the social dynamics looked uncannily familiar.

Lofgren later teamed up with epidemiologist Nina Fefferman to analyse the event in a paper published in The Lancet Infectious Diseases. They argued that massively multiplayer online games might provide a new kind of laboratory: environments populated by real people who are emotionally invested in their avatars and willing to experiment, cooperate (or sabotage) when confronted with a crisis.

Researchers at the Centers for Disease Control and Prevention even contacted the game’s developer, Blizzard Entertainment, hoping to study detailed data from the outbreak, but unfortunately the epidemic had been entirely unplanned. In the rush to fix the bug, the company had not preserved the necessary logs.

Still, the accident revealed something intriguing. Traditional epidemiological models require assumptions about how people will behave. But real crises often defy those assumptions.

The global pandemic caused by COVID-19 made this painfully clear. Early models struggled to account for sudden shifts in behavior, from panic buying and conspiracy theories to uneven compliance with public health measures. Real outbreaks are messy, unpredictable and shaped as much by psychology as by biology.

Massively multiplayer online games occupy an unusual middle ground between theory and reality. They cannot replicate real disease dynamics, but they can simulate crises involving millions of decision-making humans interacting in real time.

In 2020 volunteers from the Elysium Project recreated the phenomenon in a legacy server event called “Pandemic in Azeroth”, designed to mimic elements of COVID-19 with interesting results. When no containment measures were introduced, roughly 88% of players became infected. When sanitation and isolation measures were implemented, infections fell to about 42%.

Despite early excitement, however, the idea of using online games as epidemiological laboratories never fully took hold in the academic community. 

Back in 2005 the solution was in the end quite simple. After multiple failed attempts at controlling the outbreak, Blizzard ended it the only way it could: by resetting the game servers and issuing a patch that prevented pets from carrying Corrupted Blood outside the dungeon. Cities repopulated, skeletons vanished and life in World of Warcraft returned to normal.

But the episode left a lasting impression on scientists thinking about the relationship between disease and human behavior. The Corrupted Blood incident suggested that the most revealing experiments are sometimes the ones nobody planned.

Two decades later, the central question remains: how to inquire the difference between model predictions and actual reaction of people when a deadly disease spreads? 

A twenty-year-old software bug helped frame that question. The COVID pandemic sharpened it. But the question lingers around.

If you’re interested in how computational modelling and artificial intelligence are being used to support epidemiology, explore the upcoming VPH2026 Conference. 

Further readings:

• “Corrupted blood” and public health, Berkeley Scientific Journal, 2019
• Virtual plagues and real-world pandemics: reflecting on the potential for online computer role-playing games to inform real world epidemic research, Medical Humanities, 2013
• The untapped potential of virtual game worlds to shed light on real world epidemics, The Lancet Infectious Diseases, 2007
• Modeling infectious diseases dissemination through online role-playing games, Epidemiology, 2007
• Did an accidental ‘blood plague’ in World of Warcraft help scientists model COVID better? The results are in, Phys.org, 2022
• The Corrupted Blood Incident: How A Videogame Accident Eerily Predicted The COVID-19 Pandemic, IFL Science, 2005
• World of Warcraft’s Corrupted Blood Outbreak is Not a Model for COVID-19, The Daily Economy, 2020