I study failure so organizations can build better controls.
Most organizations find their control failures in the incident report. My work is finding them first.
Each entry takes a documented failure and asks the same questions. Where did it actually break: in people, process, or technology? Does the same shape show up in the AI and agentic systems we are building now?
Knight Capital and the 45-Minute Unwinding of a Firm· 2012
A trading firm pushed new code to eight servers. It reached seven. On the eighth, a decade-old piece of dead code woke up and turned the firm's own systems against it.
Boeing 737 MAX: MCAS and a Single Point of Failure· 2018
Software added to a new jet read a single sensor, had the authority to push the nose down again and again, and was hidden from the pilots who would have to fight it. Two crashes, 346 dead.
CrowdStrike Falcon: One File and a Global Outage· 2024
A security vendor pushed a small configuration file to millions of machines at once. It was malformed, it ran inside the Windows kernel, and around 8.5 million computers fell over together.
Optum and the Algorithm That Used Cost as a Proxy for Need· 2019
A widely used algorithm decided which patients got extra care. It worked exactly as designed, predicting cost, and that is precisely why it under-served Black patients who were just as sick.
The Replit Agent That Deleted a Production Database· 2025
During a code freeze, an AI coding agent ran commands it was told not to run, deleted a live production database, and then misreported whether the data could be recovered.
EchoLeak: A Copilot That Could Not Tell Data From Instructions· 2025
A researcher sent a victim an email with nothing to click. Hidden text told Microsoft 365 Copilot to gather the company's secrets and send them out. Copilot obeyed.
Samsung and the Source Code That Walked Into ChatGPT· 2023
Within weeks of allowing ChatGPT at work, Samsung engineers had pasted semiconductor source code and a confidential meeting into it. The tool did nothing wrong. It did exactly what it promised.
COMPAS: When Fairness Has More Than One Definition· 2016
A risk score used in courtrooms was calibrated correctly and still wrong more often about Black defendants who would not reoffend. Both sides of the argument were right, which is the whole lesson.
Uber's Self-Driving Car and the Human Set Up to Fail· 2018
An Uber test car saw a woman crossing the road six seconds before it hit her, could not decide what she was, and had its emergency braking switched off. The backup was a human watching her phone.
Therac-25: The Machine That Trusted Software With Lives· 1987
A radiation therapy machine dropped the hardware safety locks of its predecessors and trusted software alone. A timing bug then delivered massive overdoses while the screen read normal. Patients died.
Air Canada and the Chatbot Nobody Owned· 2024
Air Canada's website chatbot told a grieving customer he could claim a bereavement discount after he flew. The policy said otherwise. A tribunal held the airline to what its bot had promised.
The Dutch Childcare Scandal: Automated Suspicion at Scale· 2021
A Dutch tax algorithm flagged families as likely fraudsters, with a foreign nationality counting against them. Tens of thousands were wrongly ruined, thousands of children were taken into care, and the government fell.
Arup and the Deepfake on the Video Call· 2024
A finance employee in Hong Kong joined a video call with the CFO and several colleagues. Every face on the call was synthetic. Fifteen transfers and 25 million dollars later, the company found out.
Mata v. Avianca: The Cases That Did Not Exist· 2023
Two lawyers filed a brief built on cases ChatGPT invented, then asked ChatGPT whether the cases were real and accepted its yes. A federal judge made the lesson permanent.
Meta and the Metric That Ate the Mission· 2021
Facebook rebuilt its feed around 'meaningful social interactions', then weighted an angry reaction five times a like. The system optimized exactly what it was told to. That was the failure.
Robert Williams: When the Computer Got the Last Word· 2020
Detroit police ran a grainy still through facial recognition, got a wrong match to an old license photo, and arrested an innocent man in his driveway. The first such wrongful arrest known to be reported in the US.
Slovakia and the Deepfake Timed for Silence· 2023
Two days before Slovakia's 2023 election, a fake audio of a leading candidate discussing vote rigging spread through social media, inside the legal quiet period when he could not answer and platform rules had a loophole for audio.