The DP-3T algorithm for contact tracing (with Nicky Case)

Digital contact tracing aims to stop COVID-19 transmission during the window when people are contagious but not yet showing symptoms. Widespread testing and fast self-isolation help once symptoms appear, but the virus still spreads through contacts made while someone is asymptomatic. The core challenge is doing something faster than interviews—without turning public health into mass surveillance.

DP-3T-style approaches address that tension by using cryptography to alert people who were near an infected person, while keeping location and identity private. Instead of relying on GPS or phone geolocation histories, each phone periodically broadcasts short, uniquely pseudo-random “gibberish” messages over Bluetooth. These messages are designed to reveal nothing about who is transmitting them and nothing about where they are—no GPS data, no personal identifiers. The uniqueness matters: if two phones stay close long enough, they exchange and record each other’s random identifiers.

In the example, Alice and Bob both run compatible privacy-first tracing apps. Every five minutes, Alice’s phone emits pseudo-random strings to nearby devices and simultaneously listens for similar strings from others. If Alice and Bob remain close for more than five minutes, their phones exchange the relevant random identifiers and store what they sent and received for the last 14 days. Because the identifiers are random and contain no location or identity information, Alice’s privacy is protected from Bob, and Bob’s privacy is protected from Alice.

When Alice later develops symptoms and tests positive, her app uploads only the random identifiers she has broadcast to a hospital database. The upload uses a one-time passcode provided by her doctor to prevent spam and unauthorized submissions. Crucially, the hospital stores only meaningless random strings: it learns nothing about Alice’s identity, where she was, who she met, or even how many people she encountered. The hospital’s role is limited to distributing a list of identifiers associated with confirmed cases.

Bob’s phone periodically checks that hospital list. If Bob’s phone finds that it has previously seen any of the newly reported random identifiers within the last 14 days, it concludes that Bob was in contact with someone who tested positive and can warn him to self-quarantine. That warning interrupts further transmission chains before the virus can spread widely.

The takeaway is that contact tracing can be “one step ahead” of the virus by alerting recent contacts based on cryptographic identifiers rather than tracking real-world location histories. Open code and the ability to verify behavior are presented as key to building trust without requiring users to blindly accept surveillance claims.