Skip to main content
Promotional banner for the pentest readiness checklist
Valid Credentials Beat Your Alerts: 5 Vendor Access MythsIncident Management
4 min readFor Third-Party Governance Teams

Valid Credentials Beat Your Alerts: 5 Vendor Access Myths

Uber recently notified around 150,000 drivers that their personal information was stolen. The breach occurred through a third-party services provider, Fireworks AI, whose compromised credentials allowed attackers to access data Uber had shared with it. Despite having alerts enabled, Uber's Cloud Security team didn't detect the breach because the attackers used valid credentials.

This should concern every Third-Party Risk Management (TPRM) team. The controls were in place but bypassed by what appeared to be normal activity. Many vendor access programs rely on questionnaires and perimeter assumptions, not on understanding how stolen credentials behave. Here's where the thinking goes wrong.

Myth 1: "If alerts are enabled, we'll know when a vendor account is abused."

Reality: Alerts detect what they're programmed to detect. Most are set for failed logins, impossible travel, malware signatures, or known-bad IPs. A valid credential from an unfamiliar location can look like routine access.

In Uber's case, attackers used compromised credentials on October 7, but Uber only learned of the incident on October 9. Alerts were active, yet nothing triggered.

Detection for vendor accounts needs behavioral baselines, not just rule matches. Determine what a normal week looks like for each vendor identity: which datasets it accesses, the volume it pulls, and its operating hours. Then alert on deviations. An account that usually accesses a narrow slice of data but suddenly pulls driver's license images is a signal, even if authentication succeeded.

Myth 2: "A vendor's credentials only expose what the vendor needs."

Reality: They expose whatever the vendor was granted, which is often more than necessary. Uber stated that the compromised credentials didn't provide access to other Uber systems. That containment is a success due to scoping. However, the data within the shared environment was still sensitive: images of driver's licenses, names, addresses, dates of birth, driver's license numbers, and government-issued ID numbers.

Scoping to a system isn't the same as scoping to a purpose. Your team should test two questions for every vendor with data access. Does the vendor need identity documents, or would tokenized or masked data suffice? And does the vendor need standing access, or can access be time-boxed and approved per task? Data minimization is a TPRM control, not just a privacy one.

Myth 3: "The vendor's own incident reporting tells us the real scope."

Reality: Initial vendor statements are often revised. Fireworks initially claimed no evidence of accessed personal information, models, or datasets. On November 9, the company acknowledged the breach also affected account and billing information, as well as some customer datasets and incomplete models.

Your contract and process should assume the first statement is provisional. Build the following into your Vendor Breach Management approach:

  • Contractual Notification Timelines that require preliminary notice quickly, followed by scheduled updates.
  • A right to receive forensic findings, not just summaries.
  • Your own parallel investigation of what you shared with the vendor, based on your data inventory rather than the vendor's conclusions.

Uber's approach points the right way. Its investigation determined who accessed what from its own side of the connection.

Myth 4: "Credential hygiene is the vendor's problem."

Reality: The vendor owns its security posture, but you own the consequences. Uber filed breach notifications with multiple US states, including California, Iowa, Massachusetts, South Carolina, Texas, and Vermont, and reportedly told the Texas Attorney General's Office that 149,346 people were affected. It's providing free credit monitoring and fraud assistance. Uber, not the vendor, faced the notification burden and the relationship with the affected drivers.

Regulatory frameworks reflect this. SR 23-4 and the EBA Guidelines on Sound Management of Third-Party Risk both treat accountability as non-transferable. ISO 27036 frames supplier access controls as a shared responsibility to be specified in the relationship. So write credential expectations into the agreement: multi-factor authentication on all accounts with access to your data, rotation requirements, prompt Access Revocation for departed staff, and a Right to Audit those controls. Then verify them during Continuous Monitoring of Active Arrangements, not only at Vendor Intake.

Myth 5: "A fast response means the incident is under control."

Reality: Speed matters, but it's only one factor. Uber severed its data connections immediately after learning of the incident and launched an investigation, which likely limited exposure. By October 13, the investigation had traced the intrusion to the compromised credentials. That is a strong response.

Still, the gap between the October 7 access and the October 9 discovery is the part no response plan can fix after the fact. Fast containment shrinks the damage; it doesn't replace earlier detection. Treat response speed and detection speed as separate metrics. Measure how long a vendor-originated intrusion can run before you notice it, not just how quickly you act once you do.

What to do instead

Move from questionnaire-based comfort to control-based evidence. Concretely:

  1. Inventory vendor identities. List every credential, service account, and API key each vendor holds, and who owns each on your side.
  2. Apply Criticality Classification to data, not just vendors. Any vendor holding identity documents or government ID numbers belongs in your highest tier regardless of contract size.
  3. Reduce standing access. Prefer time-limited, task-scoped access and masked data wherever the work allows.
  4. Baseline vendor behavior. Add behavioral detection for vendor identities, since rule-based alerts won't catch valid-credential misuse.
  5. Pre-build your severance playbook. Know in advance how to cut a vendor's data connections within hours, as Uber did, and rehearse it.
  6. Contract for evolving facts. Require staged notifications and forensic disclosure, and plan on revisions.

The lesson from this incident isn't that alerts are useless. It's that an alert tuned for intruders will miss an attacker who arrives with the vendor's own keys. Design your monitoring for that case.

Application Security Isn’t Optional Anymore.

You Might Also Like