// ai security

Uber AI Vendor Breach Report: Stolen Provider Credentials Hit 149,346 Drivers

Uber AI vendor breach report: 149,346 driver records accessed through a third-party provider account

Uber is notifying 149,346 drivers that an outsider used a service provider's stolen internal credentials to access their names, driver's license numbers, dates of birth and addresses. Uber has not named the provider. TechCrunch reports it is AI infrastructure company Fireworks, which would place the Uber AI vendor breach alongside limited data theft at Notion and backdoor malware on some Cursor users' devices.

Key takeaways

  • Uber is notifying 149,346 drivers; the count comes from a Maine Attorney General filing, not the November 10 letters.
  • A provider's stolen internal credentials reached Uber's account; Uber was told October 9 and confirmed data access October 13.
  • TechCrunch links the provider to Fireworks; Uber has not confirmed it, and reported dates conflict.
  • The same reported incident hit Notion with data theft and some Cursor users with backdoor malware.

What did Uber confirm, and how did a vendor's credentials reach its data?

The 149,346 figure comes from Uber’s filing with the Maine Office of the Attorney General, not from Uber’s letters. The letters, sent November 10, 2026, describe the third party only as a service provider and give no total. The exposed fields are names, driver’s license numbers, dates of birth and addresses. That set will pass plenty of weak knowledge-based checks. The reporting does not say whether Uber offers credit monitoring or whether drivers outside the US were affected.

The attacker never needed an Uber password. Uber says the provider reported on October 9 that an unauthorized party had compromised the provider’s internal credentials and used them to access Uber’s account there. Uber severed data connections, opened an investigation, and on October 13 confirmed the intruder had reached Uber data connected to the provider. If those credentials alone did the work, as Uber’s description suggests, Uber’s SSO and MFA never got a vote.

Severing the integration stops new traffic and little else. It does not show when access began, whether API keys, OAuth grants or support sessions were also exposed, or how much data left before October 9. Those answers sit in the provider’s audit logs, and none have been published.

What does the Fireworks link prove about the Uber AI vendor breach?

Less than the headlines suggest. Uber has not named Fireworks, and no Fireworks statement about Uber appears in the reporting. TechCrunch first reported the connection and BleepingComputer relayed it. One report says Fireworks was hacked after a compromise at an unnamed third-party cloud provider.

According to the reporting, Fireworks blames a financially motivated cybercriminal group linked to a series of LLMjacking attacks, which usually means using stolen cloud or API credentials to burn or resell someone else’s model compute. The label describes a business model, not how Uber’s driver file was read.

The dates also clash. BleepingComputer places the Fireworks breach in early October, then cites TechCrunch calling it an October 24 incident, which falls after the provider’s October 9 warning to Uber. Treat the intrusion window as unknown.

Why didn't Uber, Notion and Cursor see the same damage?

Each company used the vendor differently. Notion reported data theft from a limited number of customers. Cursor reported the same, plus unnamed backdoor malware on some users’ devices. Uber’s confirmed impact is data access through its provider account, with no reported malware on drivers’ devices.

So a vendor’s assurance that its platform was not breached tells you little. What matters is which of your assets its identities could touch: stored data, API tokens, or code running on your machines.

Cursor users on implicated machines should follow Cursor’s incident guidance, then treat every secret those machines held as burned: rotate Git, cloud and API tokens, reimage rather than clean, and check provider audit logs for use of the old credentials. Our breakdown of how AI coding agents leak secrets lists where those secrets usually hide.

What should an AI-vendor access test cover?

Start with why your data sits in the vendor’s account at all. Ask each training or inference vendor how long it keeps uploaded datasets after a project ends, and which of its own identities can read them.

When we scope third-party integrations, the inventory is usually the finding. Teams know the API key and forget the OAuth app, the service account, the webhook signing secret and the vendor support role with read access. A useful test checks:

  • What the vendor’s own identities can read in your tenant, and whether you can see those reads.
  • Whether severing the connection actually works: test that the old credentials fail.
  • Whether bulk exports and object reads raise alerts, since a stolen valid credential looks like a normal login.
  • Whether logs are exported before rotation erases the session trail.

An AI supply chain security audit builds that inventory, and API penetration testing proves what each credential can reach.

In September 2022, Lapsus$ breached Uber with an MFA-fatigue attack on an Uber EXT contractor’s account. That December, data stolen from vendor Teqtivity leaked. A contractor, a vendor, and now a vendor’s credentials: one recurring entry point outside Uber’s direct control.

FAQ

How many Uber drivers were affected by the November 2026 breach?

149,346, according to Uber's filing with the Maine Office of the Attorney General. The notification letters gave no total.

Did Uber confirm Fireworks as the service provider?

No. Uber refers only to a service provider. TechCrunch reported the Fireworks link.

Does LLMjacking mean the attackers hijacked Uber's models?

No. LLMjacking typically means abusing stolen model-service credentials. It is not the described mechanism for reading Uber's driver data.

If your training or inference vendors hold customer data, we can map every identity that reaches it and prove what a stolen one can read. See our AI supply chain security audit.


// get started

Work with AgentOffense

Tell us about your target and goals. We’ll reply with scope and a fixed-price quote — usually within one business day.

./request_engagement