Knight Capital, Zillow, and a Replit AI agent: three cases where a company's own automated system destroyed the company's own money, no regulator and no plaintiff anywhere in the story. The bill scales with the authority delegated, not the intelligence of the system.
There is a whole genre of "what AI costs companies": fines, lawsuits, regulators, courtrooms. We've covered that side, the Metas and Clearviews, where someone else takes the money. This list is the other half, and it's the half that keeps operators like us up at night: cases where a company's own automated system destroyed the company's own money, with no regulator and no plaintiff anywhere in the story.
Nobody sued. Nobody fined. The system they built, trusted, and switched on simply spent it.
Three anchor cases, in escalating order of autonomy, from a dumb deploy bug, to an algorithm with a checkbook, to an AI agent that ignored instructions and then misrepresented what it had done. Every dollar figure below is taken from the company's own filings or on-the-record statements, and we state exactly what each number is (a trading loss is not a fine; a write-down is not a lawsuit), because sloppy referents are how these stories get retold wrong.
