Skip to main content
Promotional banner for a CUI buildout guide
Valid Credentials, Zero Alerts: Fixing Third-Party MonitoringSupplier & Third-Party
5 min readFor GRC Practitioners

Valid Credentials, Zero Alerts: Fixing Third-Party Monitoring

Uber had alerts set up to catch unauthorized access to its environment at Fireworks AI. Attackers used Fireworks' internal credentials, so, as Uber reported, "no alerts were triggered." Nearly 150,000 people had personal data exposed, including images of driver's licenses.

If your team manages a third-party risk program under SOC 2 or ISO/IEC 27001, this should feel uncomfortably familiar. Most programs check that a monitoring control exists. Far fewer check whether it can detect the attack that actually happens.

Why These Mistakes Keep Happening

Third-party risk programs often focus on artifacts: questionnaires, attestation reports, signed contracts. These are easy to collect and show an auditor. Behavioral visibility into what a supplier's identity is doing with your data is harder, so it gets deferred.

There's also an assumption in most detection logic: a valid credential equals a legitimate actor. This holds until the supplier itself is compromised. Then the attacker inherits every bit of trust you extended.

The mistakes below show what that gap looks like in practice.

Mistake 1: Monitoring for Bad Credentials Instead of Bad Behavior

Why it happens: Alert rules are easiest to write around authentication events: failed logins, impossible travel, unknown accounts. These catch credential stuffing but don't catch a trusted identity doing something unusual.

Consequence: In the Fireworks incident, attackers accessed Uber data on October 7. Uber wasn't notified until October 9, and the notification came from the provider, not from Uber's own detection.

Fix: Baseline what each supplier identity normally does, then alert on deviation. Look at the volume of records retrieved, access to data types the integration has never touched, time-of-day patterns, and source IP ranges versus the supplier's known egress. Map this to ISO/IEC 27001:2022 Annex A control 8.16 (monitoring activities), which expects anomalous behavior to be evaluated, and to SOC 2 CC7.2, which covers monitoring for anomalies indicative of malicious acts.

Mistake 2: Treating the Supplier's Security Report as Proof of Your Protection

Why it happens: A clean SOC 2 report feels like closure. It's a point-in-time or period-based statement about the supplier's controls, and reviewers often skim past the exceptions and the complementary user entity controls.

Consequence: The report says nothing about whether your tenant at that supplier is monitored, scoped, or segmented. Fireworks' own incident affected multiple customers, including Notion and Cursor, none of whom were connected to one another. A shared platform means a shared blast radius.

Fix: Build a review step that extracts the complementary user entity controls from every critical supplier report and assigns each one to a named owner on your side. Under ISO/IEC 27001 control 5.22 (monitoring, review and change management of supplier services), you need evidence of ongoing review, not a filed PDF. Guidance in ISO/IEC 27036 on supplier relationships helps structure what you should be asking for.

Mistake 3: Granting Data Access Wider Than the Integration Needs

Why it happens: Integrations get set up under deadline pressure. Broad scopes are faster than designing least privilege, and nobody circles back.

Consequence: When a supplier's credentials are stolen, the data exposed is the data the connection could reach. In this case, that included names, home addresses, dates of birth, driver's license numbers, and license images. Uber said the stolen login details didn't give attackers access to other Uber systems, which suggests scoping limited the damage to some degree. The data that was reachable was still enough to trigger notifications in multiple states.

Fix: Inventory every supplier data connection and document what data classes it can reach versus what it needs. Cut the difference. Pay special attention to identity documents and other high-sensitivity data; ask whether a supplier needs the image at all, or only a verification result. This supports ISO/IEC 27001 control 5.15 (access control) and SOC 2 CC6.1 and CC6.3.

Mistake 4: Having No Tested Way to Sever a Supplier Connection

Why it happens: Offboarding is planned for contract end, not for an emergency. Nobody has rehearsed pulling the plug on a live integration.

Consequence: Uber reported that it severed data connections immediately after notification. That's the right instinct, but speed depends on knowing every connection point, key, and token before the incident starts. Teams that discover integrations mid-incident lose days.

Fix: Maintain a kill-switch runbook per critical supplier: where credentials live, how to revoke them, how to disable data flows, and who can authorize it. Test it in a tabletop exercise. Tie this to your incident response procedures under ISO/IEC 27001 control 5.26 and SOC 2 CC7.4.

Mistake 5: Leaving Supplier Notification Terms Vague

Why it happens: Contracts get negotiated by legal and procurement, with security requirements added late as boilerplate.

Consequence: A two-day gap between access and notification is time attackers can use to exfiltrate and monetize. Fireworks tied the attack to a financially motivated group seeking to sell stolen AI access tokens, which is exactly the kind of activity that moves fast.

Fix: Write notification windows into supplier agreements, define what counts as a reportable event (including credential compromise, not just confirmed data loss), and require suppliers to rotate and report on credentials after any internal compromise. ISO/IEC 27001 control 5.20 (addressing information security within supplier agreements) is the anchor here.

Prevention Checklist

  • Each critical supplier connection has behavioral detection rules, not only authentication alerts.
  • Anomaly thresholds are documented per supplier and reviewed on a defined cycle.
  • Complementary user entity controls from supplier reports are mapped to named internal owners.
  • Data access per integration is documented against actual need, with excess scope removed.
  • High-sensitivity data such as identity document images is minimized or excluded from supplier flows where possible.
  • A tested runbook exists to revoke credentials and sever data connections within a defined time.
  • Supplier contracts specify notification timelines and include credential compromise as a reportable event.
  • Supplier incidents are fed back into your risk assessment and Risk Treatment Plan.
  • Tabletop exercises include a "trusted supplier is the attacker" scenario.

The uncomfortable lesson is that your alerts can be configured, enabled, and silent while the breach is underway. Audit your detection against the scenario where the credential is valid and the user is not who you think.

Promotional banner highlighting failures found in PCI audits and how to spot the gaps

You Might Also Like