Skip to main content
ACTIVELY MONITORED//Verified: Nov 16, 2026
Data BreachesCloud DefenseIdentity & Access
5 min read

Uber Data Breach: 150K Drivers Exposed via Fireworks AI

•By Rutvij Khatri•VERIFIED INTEL
Uber Data Breach: 150K Drivers Exposed via Fireworks AI

SOC Briefing Summary :: Executive Key Takeaways

  • [01]Attackers compromised credentials of AI platform Fireworks AI on October 7 to access an Uber-managed cloud environment.
  • [02]Breach impacted 149,346 drivers, exposing driver license scans, government IDs, names, addresses, and dates of birth.
  • [03]Cloud security alerts failed to trigger because threat actors authenticated using valid vendor service credentials.

Executive Incident Briefing (2-Minute CISO Brief)

Quick Answer (At-A-Glance): Uber disclosed a major third-party cloud data breach impacting approximately 150,000 drivers across multiple US states. The security incident stemmed from compromised service credentials belonging to Fireworks AI—a cloud platform utilized for running and fine-tuning open-source artificial intelligence models. Threat actors used these valid credentials on October 7 to infiltrate an Uber-managed cloud storage repository, exfiltrating images of driver's licenses, Social Security numbers, dates of birth, and tax identifiers. While security monitoring was enabled, no alerts fired because legitimate vendor credentials were authenticated. Security leaders must immediately re-architect third-party IAM trust, enforce strict IP binding, and deploy behavioral KQL queries to hunt anomalous vendor token usage.

On October 9, Uber detected unauthorized access into an isolated cloud environment managed by its Cloud Security team and immediately severed data connections with the external service provider. Subsequent forensic investigation confirmed that on October 7, adversaries utilized compromised service credentials belonging to Fireworks AI to access sensitive driver onboarding databases.

  • Incident Classification: Third-Party Vendor Cloud Data Breach
  • Affected Population: 149,346 commercial drivers across California, Texas, Massachusetts, Iowa, and Vermont.
  • Compromised Data Records: Digital images of driver's licenses, government ID numbers, tax identifiers, full legal names, physical addresses, and birth dates.
  • Third-Party Service Provider: Fireworks AI (cloud infrastructure provider for AI model hosting and fine-tuning).

The incident underscores the urgent challenge of identity-based supply chain risk: when adversaries obtain valid vendor service principal tokens, traditional signature-based detection mechanisms fail.

Technical Vulnerability Analysis & Attack Chain

Attack Chain Flow
// Attack Chain Flow

The breach architecture illustrates how modern enterprise AI integrations can inadvertently introduce lateral exposure into highly sensitive corporate data silos.

1. Ingress via Compromised AI Service Provider Credentials

Fireworks AI provides infrastructure for enterprises to deploy and fine-tune open-weight LLMs. Threat actors compromised service account credentials or API secrets associated with Fireworks AI's administrative plane. Notably, Fireworks AI later confirmed that its own systems suffered unauthorized access affecting account billing information, customer datasets, and incomplete models.

2. Cloud IAM Authentication & Alert Evasion

Equipped with valid service provider credentials, the threat actor authenticated to the Uber cloud environment on October 7. Because the authentication request utilized cryptographically valid API keys and matching service principal definitions, automated security alerts were suppressed. The adversary operated undetected for approximately 48 hours until Uber's security operations detected behavioral anomalies on October 9.

3. Enumeration of Driver Identity Data Stores

The adversary enumerated cloud object storage buckets (Amazon S3 / Azure Blob) containing driver compliance documents. The threat actor systematically extracted:

  • High-resolution scans and photographs of state-issued driver's licenses.
  • Driver licensing numbers, Social Security numbers, and state registration identifiers.
  • Linked personal profile records (home addresses, contact information, date of birth).

4. Forensic Severance & Blast Radius

On October 9, Uber revoked all access tokens, rotated cloud secrets, and severed network connectivity to Fireworks AI. Uber confirmed that the compromised credentials did not grant access to core ride-hailing production systems, passenger trip logs, or corporate financial systems. Notifications have since been submitted to the Texas Attorney General's Office and respective state regulators covering 149,346 affected individuals.

MITRE ATT&CK Tactics, Techniques & Procedures (TTPs)

MITRE ATT&CK • OPERATIONAL TTP MAPPING
TacticTechnique IDTechnique NameOperational Context
Initial AccessT1199Trusted RelationshipCompromise of third-party AI provider (Fireworks AI) credentials to access Uber cloud environment.
Defense EvasionT1078.004Cloud AccountsUse of legitimate service provider credentials bypassed automated cloud alerting rules.
DiscoveryT1580Cloud Infrastructure DiscoveryAdversary enumerated cloud storage containers hosting driver onboarding records.
CollectionT1530Data from Cloud StorageAccess and extraction of over 149,000 driver's license scans and personal identification records.
ExfiltrationT1567.002Exfiltration to Cloud StorageBulk data exfiltration over encrypted HTTPS channels to external adversary storage.

Detection, Hunting & SOC Mitigation Playbook

1. Immediate Containment Checklist

  • [ ] Audit all third-party AI and SaaS vendor integrations; revoke and rotate any API keys and service principal credentials older than 90 days.
  • [ ] Enforce conditional access policies mandating IP whitelisting / IP binding for all external vendor service accounts.
  • [ ] Enable Object-Level Access Logging (e.g., AWS S3 Server Access Logging / CloudTrail Data Events) on all buckets housing sensitive PII.
  • [ ] Terminate active OAuth and service principal tokens associated with unverified foreign ASN logins.

2. Behavioral Detection Strategy

When monitoring vendor service accounts:

  1. Establish baseline access profiles for third-party service principals: standard access hours, query frequencies, and geographic IP origins.
  2. Flag any high-volume GetObject or BlobDownload operations where the daily retrieval rate deviates by more than 3 standard deviations from normal baseline.
  3. Monitor for cross-tenant token usage or API requests presenting novel User-Agent strings.

3. Endpoint Detection & Hunting Query

Sigma Detection Rule

QUERY / DETECTION_RULE
SIGMA / YAML
title: Anomalous Third-Party Service Principal Sign-In from Foreign ASN
id: c714e1a0-529a-41d3-9821-2a13b4c9101f
status: experimental
description: Detects service principal or vendor service account authentication originating from unfamiliar IP addresses or ASN networks.
author: Rutvij Khatri (CyberNewsAI)
references:
    - https://www.securityweek.com/uber-data-breach-impacts-150000-people/?ref=cybernewsai.com
date: 2026-11-16
logsource:
    category: identity
    product: azure
detection:
    selection:
        AppDisplayName|contains:
            - 'Vendor'
            - 'AI'
            - 'Fireworks'
            - 'Service'
        ResultType: 0
    condition: selection
falsepositives:
    - Vendor infrastructure dynamic IP migrations or new region deployments.
level: medium
tags:
    - attack.initial_access
    - attack.t1199
    - attack.t1078.004

Microsoft Sentinel / Defender Detection KQL (Anomalous Vendor Sign-Ins)

QUERY / DETECTION_RULE
SENTINEL / KQL
// Detection Query: Hunt for third-party vendor service principal authentications from unexpected IP origins
AADServicePrincipalSignInLogs
| where Timestamp >= ago(7d)
| where ServicePrincipalName has_any ("Fireworks", "Vendor", "AI-Integration", "ExternalService")
| extend IPAddress = tostring(IPAddress)
| extend Location = tostring(LocationDetails.countryOrRegion)
| summarize SigninCount = count(), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by ServicePrincipalName, IPAddress, Location, UserAgent
| where SigninCount < 10 or Location !in ("US", "CA")
| sort by SigninCount asc

Microsoft Sentinel / Defender Lateral Movement & Hunting KQL (Cloud Storage Exfiltration)

QUERY / DETECTION_RULE
SENTINEL / KQL
// Lateral Movement / Exfiltration Hunting: Detect bulk data download events from sensitive cloud storage by service accounts
CloudAppEvents
| where Timestamp >= ago(7d)
| where ActionType in ("GetObject", "DownloadFile", "BlobDownload")
| where AccountType =~ "ServicePrincipal" or AccountName has "Vendor"
| where RawEventData has_any("driver", "license", "identity", "pii", "tax")
| summarize DownloadCount = count(), TotalBytes = sum(tolong(RawEventData.bytesReceived)) by AccountName, IPAddress, tostring(RawEventData.bucketName), bin(Timestamp, 1h)
| where DownloadCount > 500
| sort by DownloadCount desc

4. Enterprise Hardening & Long-Term Resilience

  1. Zero Standing Privileges (ZSP): Transition third-party integrations to Just-In-Time (JIT) access tokens with lifespans strictly limited to 1 hour.
  2. Data-Centric Tokenization: Store sensitive identity artifacts (such as driver's licenses) in format-preserving tokenized vaults where decryption keys require multi-party cryptographic authorization.
  3. Continuous Vendor Telemetry Auditing: Mandate that third-party AI vendors provide automated SOC2 Type II compliance feeds and real-time security telemetry integration.

Indicators of Compromise (IOCs)

Indicator TypeValueForensic Description
other149,346 driver records, driver license scans, government IDsForensic blast radius of compromised driver records
otherFireworks AI cloud model fine-tuning platformThird-party vendor whose compromised credentials enabled cloud ingress

---

Related Topic Cluster Intelligence

Indicators of Compromise (IOCs)
2 Identified
BREACH BLAST RADIUS149,346 driver records, driver license scans, government IDs
COMPROMISED VENDORFireworks AI cloud model fine-tuning platform

// VERIFIED_SOURCES_&_REFERENCES

SHARE INTEL:Reddit