The Uber data breach exposed driver's license images held in a cloud environment built for an AI vendor. What it means for vendor oversight and ID retention.
Table of contents
The Uber data breach exposed images of drivers' licenses, plus the text extracted from them, that Uber had placed in a cloud environment built for a service provider. An attacker used the provider's stolen credentials on 7 October 2026. Uber's Oregon filing lists 149,346 people affected, and Uber has identified the provider as Fireworks AI.
- Uber's notice to the Iowa Attorney General says the data included images of driver's licenses and extracted name, address, date of birth and license number.
- The data sat in a cloud storage environment Uber created to enable the provider's work. It had alerting enabled, but valid credentials meant no alert fired.
- Uber's filing on the Oregon Department of Justice breach register lists 149,346 people affected. Iowa's notice covers about 2,999 residents, and the enclosed notice template puts Rhode Island at about 1,873.
- Fireworks AI says its October incident was linked to a financially motivated group selling stolen access to large language models.
- Cursor reported the same pattern: an access key it issued to Fireworks was used to copy a storage bucket shared with the vendor.
What happened in the Uber data breach?
The Uber data breach is a third-party credential compromise, not an intrusion into Uber's core systems. Uber says an unauthorised party stole a service provider's internal credentials and used them to reach Uber's account with that provider. The data exposed was driver's license images and their extracted text, stored for the provider's work.
Uber set out the sequence in a notification letter to the Iowa Attorney General dated 10 November 2026. The provider told Uber on 9 October that its credentials had been compromised. Uber says it "immediately severed data connections" and opened an investigation. On 13 October it found that the credentials had been used on 7 October to access Uber data connected to the provider.
The letter describes the data as "images of driver's licenses and extracted text data, including name, address, date of birth, and driver's license number." The same date of breach, 7 October 2026, appears in Uber's filing on the California Attorney General's breach register, which publishes the multistate notice template sent to individuals. The Oregon register lists 1 October 2026 as the breach date, and Uber has not explained the difference.
Uber's letters do not name the provider. An Uber spokesperson told TechCrunch that the notices relate to an incident at Fireworks.ai, which "had access to some Uber data."
| Item | Detail from Uber's filings |
|---|---|
| Date of access | 7 October 2026 |
| Uber notified by the provider | 9 October 2026 |
| Access confirmed by Uber | 13 October 2026 |
| Notices to individuals | From 10 November 2026 |
| Data types | License images; extracted name, address, date of birth, license number |
| People affected, Oregon register | 149,346 |
| Iowa residents | About 2,999 |
| Rhode Island residents | About 1,873 |
| Remedy offered | Credit monitoring and fraud assistance through Cyberscout |
The 149,346 figure comes from Uber's entry on the Oregon Department of Justice register, and TechCrunch reports Uber said this was the total it notified. Uber has not published a breakdown by state beyond the individual filings.
Why did no alert fire?
No alert fired because the attacker used credentials that were valid for the environment. Uber's letter says the storage environment was overseen by its Cloud Security team and "had alerting enabled." It cites "valid credentials and data flows enabled by the compromise" of the provider's system. That suggests the access resembled normal vendor activity.
The key sentence in Uber's notice is short: "no alerts were triggered." Alerting tuned to spot unauthorised access cannot easily tell a stolen vendor key from the vendor itself. Uber also says the compromised credentials did not reach any other Uber system, which suggests the environment was scoped to one purpose. That limited the spread, but not the loss of what was inside.
The Fireworks side of the chain is described in its own security bulletin update, last updated 9 November 2026. Fireworks says it first published a bulletin on 9 October, brought in Palo Alto Networks Unit 42, and later found impacts to account and billing information and, for a limited set of accounts, to "secrets, datasets, and incomplete models." It links the activity to a group selling stolen access to large language models, a practice known as LLMJacking. Fireworks also reports a second intrusion: a third-party cloud provider hosting its R&D machines was compromised on 23 October, and the attacker used it on 24 October to reach Fireworks development machines.
Cursor's incident report shows the same mechanism from a second customer. Cursor says an attacker obtained an access key Cursor had issued to Fireworks and used it to copy a read-only storage bucket "we used to share data with them." In both cases the customer built a store for the vendor, and the vendor's credential opened it. TechCrunch reports that Notion also confirmed data was taken in the Fireworks incident.
| Company | What was shared with the vendor | How it was reached |
|---|---|---|
| Uber | Driver's license images and extracted text | Vendor's compromised credentials, valid for Uber's environment |
| Cursor | Coding agent traces in a read-only bucket | Access key Cursor issued to Fireworks |
What does the Uber data breach mean for your obligations?
The Uber data breach maps onto three duties regulated firms already carry: vendor oversight, data minimisation in identity records, and breach notification. None of them is new. What is new is the route: Uber and Cursor both built data stores for the same AI vendor, and its compromise exposed both.
Vendor oversight. For non-bank financial institutions under the FTC Safeguards Rule, 16 CFR 314.4(f) requires selecting capable service providers, contractual safeguards, and "periodically assessing your service providers based on the risk they present" and the continued adequacy of their safeguards. New York's 23 NYCRR 500.11 requires written policies for nonpublic information held by third-party service providers, covering access controls, multi-factor authentication and encryption. An AI vendor processing ID images belongs in that inventory, with its own risk rating.
Identity record contents. For banks, the CIP recordkeeping rule at 31 CFR 1020.220(a)(3) requires a description of the document relied on: its type, number, place of issuance, and issue and expiry dates. It does not require a stored image. If a vendor task needs only extracted fields, the image does not have to travel. See our CIP glossary entry for the full record list.
Breach notification. State breach laws set the individual notices Uber is now sending. Firms regulated by New York's DFS must notify the superintendent within 72 hours of determining that a cybersecurity incident occurred, including one at a third-party service provider, under 23 NYCRR 500.17(a). Under 16 CFR 314.4(j), a notification event, defined in 16 CFR 314.2(m) as unauthorised acquisition of unencrypted customer information, involving at least 500 consumers must reach the FTC no later than 30 days after discovery. Where EU or UK GDPR applies, Article 28 requires processors "providing sufficient guarantees," and Article 33 gives the controller 72 hours from awareness. In each regime, the clock can start with the vendor's call, not your own detection.
What is still uncertain about the Uber data breach?
The main open questions are scale, purpose and onward use. Uber has not published a state-by-state breakdown, said why the provider needed license images rather than extracted fields, or said how long the images were kept. Fireworks has not said how many customers it notified, and the start date differs between the Oregon and California filings.
Scale. The 149,346 total rests on Uber's Oregon filing, which TechCrunch reports Uber described as the total notified. Individual state counts, such as Iowa's 2,999, are published only where a state requires them, so the geographic spread is still incomplete.
Purpose and retention. Uber's letter says the environment was "created by Uber specifically to enable the service provider's work." It does not describe that work. Whether images were needed, and whether they were deleted after the task, decides how much was exposed. That is the question regulators and plaintiffs are likely to press.
Onward use. Uber says it has not identified misuse to date. Fireworks attributes its incident to a group that sells model access, which points to compute theft as the motive. That does not mean license images will stay unused. A license image with name, address and date of birth is enough material for synthetic identity and account opening fraud, as covered in our synthetic ID glossary entry.
Detection. The incident exposes a gap that is hard to close. Alerts tuned for unauthorised access may not catch a stolen key that behaves like the vendor it belongs to. Behavioural baselines per vendor can help, but they add cost and false positives.
How does it compare with other ID document breaches?
The Uber data breach is smaller in reported scale than the largest 2026 ID document exposures, but its route is different. Earlier breaches hit the company or vendor holding the documents directly. This one ran through a general AI provider that held identity data for a separate task.
Earlier this year, the IDScan data breach put a marketplace claim of 153 million ID scans in front of the courts. The Times Car data breach exposed identity document images for about 1.6 million accounts. The DigiCert data breach exposed validation archives that included passports and licenses.
| Incident | Holder of the images | Route in | Reported scale |
|---|---|---|---|
| Uber, October 2026 | Uber environment built for an AI vendor | Vendor's stolen credentials | 149,346 on the Oregon register |
| Times Car, September 2026 | Car-sharing operator's web system | Unauthorised access to that system | ID images for about 1.6 million accounts |
| DigiCert, June to July 2026 | Certificate authority validation archive | Self-registered outside account | Not published |
| IDScan, September 2026 | ID scanning vendor | Security incident confirmed, route not published | 153 million scans advertised, unverified |
In the three confirmed cases, and in the records advertised in the IDScan case, identity documents or their data were kept beyond the moment of verification, and that copy was what was exposed. The Uber data breach adds a vendor credential as a new way in.
How should compliance teams respond?
Start with an inventory of every place identity document images leave the verification system, the kind of onward copy at the centre of the Uber data breach. Then cut what vendors receive to the fields their task needs. Neither step needs new technology, and both reduce what a stolen vendor key can reach.
- List every vendor, including AI and analytics providers, that can read ID images or extracted identity fields, and record why.
- Replace images with extracted fields where the task allows, and set short retention for any shared store.
- Issue each vendor scoped, short-lived credentials, and baseline normal volumes so bulk reads stand out.
- Write vendor breach-call timelines into contracts so your notification clock is not lost in the hand-off.
- Test whether your CIP and CDD records hold the document description the rules require without the image.
Zyphe takes a different route. After verification, documents and biometrics are encrypted into the user's own vault, split across storage nodes and unlockable only with a key the user or customer holds. Businesses work from verification results and proofs, and reach underlying documents only through access the user consents to, so there is no central store of customer PII to hand to a vendor. See how Zyphe stores PII, read how it works, or book a demo.
The bottom line
The Uber data breach shows identity data leaking not from the verification step but from copies sent onward, here to a store built for an AI vendor. For KYC and AML teams, the lesson is to map every onward copy, share fields instead of images, and give AI providers the same oversight as any other service provider.
Cited sources
- Uber Technologies notification letter and individual notice, Iowa Attorney General, 10 November 2026
- Uber Technologies breach filing, California Attorney General register
- Oregon Department of Justice data security breach register
- Fireworks AI, update on October 2026 security incident
- Cursor, security incident update
- 31 CFR 1020.220, customer identification programs for banks
- 16 CFR 314.4, FTC Safeguards Rule elements
- 16 CFR 314.2, FTC Safeguards Rule definitions
- 23 NYCRR 500.11, third-party service provider security policy
- 23 NYCRR 500.17, notices to the superintendent
- GDPR Article 28, processor
- GDPR Article 33, notification of a personal data breach
- TechCrunch, Uber says hackers stole information on close to 150,000 driver's licenses
Michelangelo Frigo (Co-Founder at Zyphe) Michelangelo Frigo is a privacy and identity infrastructure expert and co-founder of Zyphe.