Uber is notifying drivers that hackers stole their personal information and driver's licence details from one of its vendors — and the vendor, according to reporting by TechCrunch that Uber confirmed, is Fireworks AI, the inference provider valued at $17.5 billion in July.
The disclosure is the newest fallout from a pair of October breaches at Fireworks AI, and it shows how far the blast radius of a compromised AI vendor can extend. In recent weeks the same attacker activity has been linked to a trojanized Claude Code extension that reached Cursor users, and to exposed chat transcripts at Notion. For anyone tracking the latest AI developments, the pattern is becoming hard to ignore: when AI infrastructure is breached, the damage rarely stays inside the vendor's own walls.
How the Breach Reached Uber's Drivers
Uber said Fireworks told it on October 9 that attackers had stolen internal credentials. Those credentials were then used to reach Uber's data, according to TechCrunch. Uber confirmed the vendor's identity but has not said how many people were affected in total.
What is known comes from breach notifications Uber filed with state attorneys general this week. Those filings list 2,567 affected people in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island — nearly 7,900 drivers so far, and that count covers only the three states that have disclosed figures publicly.
Two Breaches in One Month at a $17.5 Billion AI Vendor
Fireworks AI sells developers access to open AI models, handling inference workloads for enterprise customers. It has disclosed two separate incidents from October. In the first, intruders misused internal credentials. In the second, they reached the company's development machines through a breach at a cloud provider.
According to Fireworks, the attackers took billing data. For some customers, they also took secrets, datasets and unfinished model weights. Notion, one of Fireworks' customers, has said that AI chat transcripts from fewer than 1% of its workspaces were exposed in the incident.
The Cursor incident, which the company detailed in a November 13 blog post, traced a malicious copy of Anthropic's Claude Code extension in its marketplace back to an access key that had been issued to Fireworks and exposed there. Up to 5,000 users were estimated to have been affected during the exposure window.
LLMjacking: Stolen AI Credentials as a Product
Fireworks has blamed a group that sells stolen access to AI models for profit. The practice is known as LLMjacking: rather than extorting victims with ransomware, attackers harvest valid credentials and quietly resell access to powerful models, letting other users generate AI output at the victim's expense.
The economics explain the persistence of these campaigns. Inference capacity for frontier-class models is expensive to buy legitimately, so a working set of stolen credentials functions like inventory in a criminal marketplace. No group has claimed responsibility for the Fireworks theft.
The Open Question: Why Did Licence Data Sit With an Inference Vendor?
Perhaps the most striking detail in the disclosure is one neither company has explained: why driver's licence data was held by an AI inference provider at all. Uber referred other questions to Fireworks, which did not answer TechCrunch's specific questions.
The incident is Uber's second known breach this year, following an Uber Freight incident in August. That frequency matters for a company whose core business depends on drivers trusting it with identity documents.
What Is Known — and What Isn't
Several basic facts remain undisclosed. Uber has not given a total number of affected drivers, and it is not publicly known whether data beyond licence details — such as Social Security numbers or bank information — was reached in the Uber-specific exposure. Drivers are being notified directly, and the state-by-state filings are how the affected count is being assembled.
Fireworks, for its part, has said the attackers took billing data from customers, along with secrets, datasets and unfinished model weights in some cases. Whether the licence data was part of a dataset used for a customer workload, or stored for another reason, has not been stated by either company.
A Supply-Chain Pattern AI Companies Can't Ignore
Taken together, the Uber, Cursor and Notion disclosures sketch a supply-chain problem that is specific to the AI industry's current structure. Inference vendors hold three valuable things at once: customer credentials that open doors to other companies, training data that can include sensitive personal information, and model weights that represent competitive advantage.
For enterprises, the practical lesson from the October incidents is that an AI vendor's security posture is now part of their own attack surface — and that contracts with inference providers deserve the same scrutiny as any other vendor that touches customer data. For vendors, the lesson is that credentials are the crown jewels: the first domino in every one of these incidents was a set of keys that got out.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →