As reported by BleepingComputer, Uber is notifying nearly 150,000 drivers that their personal data — including names, driver's license numbers, dates of birth, and addresses — was stolen after attackers compromised AI infrastructure provider Fireworks through a third-party cloud provider breach. The incident, traced to early October 2026, illustrates a pattern we at Shield53 have been tracking with growing concern: AI/ML service providers are becoming privileged choke points in the modern supply chain, and their compromise creates downstream blast radius that most organizations haven't mapped.

Key Takeaway: As reported by BleepingComputer, Uber is notifying nearly 150,000 drivers that their personal data — including names, driver's license numbers, dates of birth, and addresses — was stolen after attackers compromised AI infrastructure provider Fireworks through a third-party cloud provider breach.

Why This Matters Beyond Uber

The Fireworks breach didn't just affect Uber. According to the reporting, the same incident propagated into security incidents at AI coding platform Cursor and workspace provider Notion — with Cursor's case involving backdoor malware deployment on user devices. This is textbook supply chain cascade: a single provider compromise generating multiple downstream incidents with materially different impact profiles (data exfiltration at Uber, malware delivery at Cursor).

The attack vector here — LLMjacking via compromised cloud credentials — represents an evolution in financially motivated cybercrime. Attackers aren't just stealing data; they're hijacking AI compute resources and leveraging the trust relationships these providers hold with their customers.

The exposed data set is particularly concerning. Driver's license numbers combined with dates of birth and addresses create a complete identity theft toolkit. For gig economy workers — a population already facing economic precarity — the downstream fraud risk is significant and long-tailed. This isn't a breach that ends when Uber severs the connection; the data is now in circulation.

The Third-Party Risk Problem Nobody Solved

Uber reportedly severed data connections with Fireworks on October 9, four days before confirming the extent of unauthorized access. That's a reasonable incident response timeline, but it highlights a structural gap: the connection existed and was actively transferring sensitive PII without sufficient isolation or monitoring. Organizations sharing PII with AI infrastructure providers should be asking why those data pipelines weren't designed with break-glass controls, rate limiting, and anomaly detection from the start.

This is also Uber's fourth publicly disclosed breach since 2016. While each incident involved different vectors and third parties, the pattern suggests that vendor risk management at scale remains an unsolved problem even for well-resourced organizations.

Shield53 Recommendations

Shield53 Recommendations
Inventory AI/ML vendor data flows: Map every third-party AI or cloud service that has access to production data. If you can't produce this inventory within 24 hours, you have a visibility problem.
Implement least-privilege API scoping: Vendor integrations should receive the minimum data fields required for the business function. Driver's license numbers should not be flowing to an AI training platform unless absolutely necessary — and if they are, they should be tokenized or hashed.
Deploy continuous vendor monitoring: Static vendor risk questionnaires are insufficient. Monitor for credential exposure, anomalous access patterns, and provider-side incidents in real time using threat intelligence feeds.
Establish break-glass procedures: Pre-define the technical steps and decision authority for severing a vendor data connection. Uber's response was appropriate but reactive; the capability to cut access should be rehearsed and ready.
Assume breach when sharing PII: Apply encryption-in-use, data loss prevention tagging, and contractual breach notification SLAs (24-48 hours, not days) to any vendor handling regulated or sensitive data.
Brief affected populations proactively: For gig workers exposed to identity theft risk, provide credit monitoring and identity restoration services — not just notification letters.

The broader industry takeaway is uncomfortable: as AI infrastructure becomes foundational to more business operations, the attack surface it creates is growing faster than the governance frameworks meant to secure it. Organizations consuming AI services need to treat these providers with the same scrutiny applied to any other critical infrastructure vendor — because attackers clearly already do.