AI Governance Institute
← News
Research2026-11-18

Fireworks AI Breach Exposes 149,000 Uber Drivers' License Data

What happened

Uber confirmed that attackers accessed driver personal data held by its AI inference supplier, Fireworks AI, in an incident first disclosed in a report by The Register. A breach notification filed with Oregon's attorney general places the number of affected individuals at 149,346. The stolen data included images of drivers' licenses and information extracted from them: names, home addresses, dates of birth, and license numbers. Fireworks AI alerted Uber on October 9; Uber cut off the supplier's access and confirmed four days later that driver data had been accessed on October 7. Uber had provisioned a dedicated cloud storage environment for Fireworks AI, with its own security monitoring, but the attackers used legitimate stolen credentials, so no alerts fired. This breach is distinct from an earlier incident at Fireworks AI that month, which is covered in Fireworks AI's Second Breach Exposes Model Weights, Billing Data via Cloud Provider. It is also separate from an incident affecting Notion, whose customers' AI assistant conversations were exposed. Uber began notifying affected drivers on November 10, more than a month after learning of the breach. Uber is offering 90 days from the notification letter to enroll in free credit monitoring.

Why it matters

  • ·Organizations sharing government-issued identity documents with AI vendors face compounded regulatory exposure. Purpose-limitation and data-minimization requirements under privacy laws such as the General Data Protection Regulation (GDPR) require firms to justify why such sensitive data is necessary for AI inference tasks at all, and to minimize what is transferred.
  • ·The credential-based intrusion bypassed Uber's dedicated monitoring environment entirely, exposing a structural gap in how enterprises scope security controls for AI sub-vendors. Perimeter and alerting tools designed for known attack patterns cannot detect access by actors using legitimate stolen credentials. Contractual and technical data-access limits therefore become the last line of defense.
  • ·The more-than-30-day gap between Uber's discovery on October 9 and driver notification on November 10 creates breach-notification compliance risk in multiple US states. Two separate incidents at the same vendor within one month signal that a prior-incident re-assessment trigger was available but apparently not activated in time to prevent data loss in this breach.

Stay ahead of stories like this

Get developments like this, plus everything else that matters in AI governance. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Audit every active AI inference, fine-tuning, or model-hosting vendor relationship to identify which ones hold copies of personal data, government-issued identity documents, or other regulated information, and confirm in writing why each category of data is necessary for the AI task.
  • ☐Review vendor contracts with AI suppliers to confirm they include a mandatory breach-notification window short enough to meet your most stringent applicable state or national breach-notification deadline, and verify those notification obligations were actually triggered and met in any recent incidents.
  • ☐Check whether your vendor risk program includes a re-assessment trigger when a supplier discloses any security incident: if Fireworks AI's first October breach was known, determine whether a formal reassessment of data access was initiated before this second breach occurred.
  • ☐Ask your cloud and security teams how credentials issued to AI suppliers are scoped, rotated, and revoked: confirm that stolen supplier credentials cannot access more data than the minimum required, and that revocation can happen within hours of an incident being reported.
  • ☐Determine whether drivers or other individuals whose government-issued identity data was shared with this vendor have been notified within the timeframe required by each applicable state's breach-notification law, and document the basis for the notification timeline chosen.

What to watch next

State attorneys general in Oregon, Iowa, and potentially other states have received breach notifications. Enforcement actions based on the notification timeline and the scope of data shared are a live risk. As more states activate their own AI and privacy enforcement programs, regulators may ask why sensitive identity documents were provided to an inference vendor rather than anonymized or pseudonymized data. This question may attract formal inquiry. Compliance teams should also monitor whether this incident triggers guidance from privacy regulators on data minimization obligations when engaging AI sub-processors. Teams should watch for any further disclosures from Fireworks AI about the total scope of data accessed across all affected customers.