Agent chaosVerified·14 Nov 2026
Uber says hackers took drivers' licence data through its AI vendor Fireworks
Uber is notifying drivers that their personal and driver's licence details were stolen in the breach at its AI vendor Fireworks. Filings in three US states alone count 7,870 people, and more states are still reporting.
The Fireworks AI breach keeps finding new victims. This week it was Uber's turn: the ride-hail company is writing to drivers to say their personal information and driver's licence details were taken from data it had shared with the AI vendor.
What happened
In notices filed with several US state attorneys general, Uber said a service provider told it on 9 October that hackers had stolen internal credentials and used them to get into Uber's account with that provider. Uber cut its data connections with the provider and started investigating. On 13 October it concluded that the intruder had used those credentials to access Uber data held there, including drivers' personal information.
Uber has since confirmed the provider is Fireworks AI, a platform that lets developers run open AI models. Fireworks confirmed unauthorized access to some of its systems that affected customer data. It had already disclosed a breach in which customers' billing data and some private AI datasets were taken, and Notion has confirmed its data was caught up too.
How big is it
State filings show driver's licence data for 2,567 people in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island. That is 7,870 people from three states. Drivers in Illinois and California were notified as well, but those counts were not public yet, so the real total is likely higher.
Uber has not said how many people are affected overall. It says it has found no misuse of the data so far, and it is offering affected drivers free credit monitoring and fraud help for 90 days from their letter.
Who did it
Fireworks has blamed a financially motivated criminal group that sells stolen access to large language models. No group has publicly claimed the theft.
Our take
Nobody's AI agent went rogue here. The fail is quieter and more common: real people's ID data was sitting with a third-party AI platform, and one stolen set of credentials there was enough to reach it. Every dataset you hand to a model vendor becomes part of that vendor's attack surface. Send the minimum, strip identity documents and personal details first, use separate short-lived credentials per vendor, and know exactly what each one holds before you need to write the breach notice.
Every dataset you hand an AI vendor is now their attack surface too, so send the minimum.
