Skip to main content
promotional banner for can you prove your CC^ checklist
Valid Credentials Beat Your Cloud Alerts: Uber's Vendor LessonIncident Management
3 min readFor Vendor Due Diligence Analysts

Valid Credentials Beat Your Cloud Alerts: Uber's Vendor Lesson

Breach Overview

Uber reported to regulators that a breach exposed names and driver's license data for 149,346 people. The breach wasn't due to a flaw in Uber's systems but occurred through a third-party service provider. On October 9, the provider informed Uber that internal credentials had been stolen and used to access Uber's account.

The data was stored in a cloud environment Uber created for the provider's work. Attackers accessed driver's license images and extracted personal information such as names, addresses, dates of birth, and license numbers.

The key takeaway for your Third-Party Risk Management (TPRM) program is that Uber's Cloud Security team managed this environment, following Uber's cloud security processes. However, no alerts were triggered because the unauthorized access used valid credentials. Even with sound detection tools, unauthorized access can go unnoticed.

Key Findings

1. Valid credentials bypass detection. Uber's investigation found that monitoring systems didn't trigger alerts because the access appeared legitimate. Detection systems focused on exploits and anomalies won't catch logins that authenticate correctly. Your team needs to focus on behavior patterns like access volume, timing, and location, not just authentication failures.

2. Purpose-built environments can still be vulnerable. Uber's dedicated cloud storage for the provider was an effective isolation strategy, limiting access to other systems. However, isolation only limits the blast radius; it doesn't protect the data within. In this case, the environment contained sensitive driver's license images.

3. Upstream compromises affect you. The incident was linked to a cyberattack on AI infrastructure company Fireworks, which reported stolen billing and AI-specific data. Although Uber didn't name its provider, the incident highlights that your data can be exposed by breaches at companies you've never assessed. This is a Fourth and Nth Party Management issue, often overlooked in pre-contractual assessments.

4. LLMJacking expands threat models. Research by Google Threat Intelligence Group and Mandiant describes LLMJacking, where attackers hijack access to large language models. Mandiant responded to multiple incidents in 2026, affecting various industries. After Fireworks disclosed its breach, customers like Notion and Cursor reported impacts. One provider's issue can cascade into multiple customer incidents, illustrating Provider Concentration Risk.

5. Rapid containment is crucial. Once notified, Uber quickly severed data connections and rotated credentials. Effective response relies on knowing every credential and connection a provider holds. If you can't identify them quickly, your Vendor Breach Management plan needs improvement.

Implications for Your Team

As a Due Diligence analyst, recognize that a vendor's clean assessment may not fully protect you. A vendor can maintain strong credential hygiene yet still have credentials stolen and used against you. Your assessments should cover what the vendor can access within your systems, not just their own security measures.

Key gaps to address:

  • Forgotten access. Vendor-specific environments often outlast projects, with credentials remaining active longer than necessary.
  • Data minimization. Evaluate whether vendors need original data like government ID images or just extracted fields, and whether they need to retain them post-processing.
  • AI vendor classification. AI infrastructure providers are often misclassified as low-risk. If they handle billing data, model access, or customer data, this classification should be revised.

Action Items by Priority

This Week

  1. Inventory all credentials, API keys, and service accounts held by third parties in your cloud environments. Document the owner, last rotation date, and scope.
  2. Ensure you can revoke access for top-tier vendors within hours. Test this with one vendor.
  3. Review your Criticality Classification for AI infrastructure and LLM providers. Reclassify those handling sensitive data or accessing model endpoints.

This Quarter

  1. Implement short-lived, scoped credentials for vendor access, with rotation in contracts. Restrict logins by network origin to prevent misuse of stolen credentials.
  2. Add behavioral monitoring for vendor-dedicated environments, focusing on bulk downloads, off-hours access, and new source locations.
  3. Update Notification Timeline clauses to require vendors to report credential compromises promptly.
  4. Request Sub-Processor Disclosure from AI and cloud vendors to map potential data exposure points, aligning with ISO 27036 and SR 23-4.

This Year

  1. Incorporate LLMJacking scenarios into tabletop exercises, including vendor breaches affecting your model access and billing.
  2. Conduct a Concentration Risk Assessment on AI infrastructure providers to identify shared critical workflows.
  3. Add data-retention and deletion verification to your audit scope for vendors handling identity documents.

EBA Outsourcing Guidelines

a promotional banner asking how ready are you for PCI DSS 4.0? With a call-to-action to get the checklist now.

You Might Also Like