In 2012 Knight Capital lost $440M in one morning because of bad code. It is one of the clearest examples of how little time it takes for a software mistake to become a financial one.
What happened
A faulty software release reached live trading in 2012. Within a single morning the firm had lost $440M.
Trading systems act in fractions of a second and at scale, so a mistake does not wait for a human to notice. By the time it is seen, the damage is done.
Where an independent check was missing
The risk sat in the release itself: how the new code was put live, whether every system received the same version and whether anyone could stop it quickly once it misbehaved.
An independent review of the release process, with a checklist signed off by someone who did not write the change, is cheap compared with $440M.
What to do differently
Release in small steps and watch each one. Use a kill switch that anyone on the team can pull, and set limits that stop the system automatically when behaviour looks wrong.
Rehearse the rollback. A team that has practised undoing a release acts in minutes rather than debating in the middle of an incident.
Match the care to the stakes
The more money or safety a system controls, the more checks its releases deserve. That applies to a trading platform, a payments flow or anything customers rely on.
Make the release process boring on purpose: repeatable, reviewed and reversible.
- One morning of bad code cost Knight Capital $440M.
- Have a second person review how a release goes live.
- Add automatic limits and a kill switch.
- Practise rollback before you need it.
