Trusted by 500+ global enterprises

Uber Data Breach via Fireworks AI: What Happened and What It Means for AI Vendor Risk

Share

Thousands of Uber drivers are opening data breach letters this month that say their driver’s license images were stolen, and the company that lost them was not Uber. Attackers Uber Data Breach via Fireworks AI: What Happened and What It Means for AI Vendor Risk

Thousands of Uber drivers are opening data breach letters this month that say their driver’s license images were stolen, and the company that lost them was not Uber. Attackers reached them through Fireworks AI, a platform Uber used to run AI models, where they took credentials in early October and used them to reach data Uber had shared with the vendor. Notion and Cursor lost data in the same Fireworks AI breach, which makes the Uber data breach one of the clearest cases yet of an AI provider becoming a shared point of failure for its customers.

Last updated November 16, 2026.

What happened in the Uber data breach?

Attackers used compromised credentials belonging to Fireworks AI, one of Uber’s technology vendors, to access Uber data the vendor held. Fireworks told Uber on October 9, 2026, and Uber’s investigation found on October 13 that the credentials had been used on October 7. Stolen records include driver’s license images and other identity data. Notices to drivers went out in November, after Uber had cut its data connections with the vendor.

Every date in this timeline comes from the companies’ own statements or public filings.

Date (2026)Event
October 6Fireworks AI detects unauthorized use of internal credentials
October 7Attackers use compromised service-provider credentials to access Uber data
October 9Fireworks notifies Uber; Uber cuts its data connections with the vendor
October 13Uber’s investigation confirms its data was accessed
October 23A third-party cloud provider hosting Fireworks research machines is breached
October 24Attackers use that access to reach Fireworks development machines
October 29Malicious copies of a coding extension appear on Cursor’s marketplace, using credentials taken from data Cursor shared with Fireworks
November 8Fireworks confirms to Notion that Notion customer data was accessed
Week of November 9Uber files breach notices with several US state attorneys general

What data was exposed, and who is affected?

Everyone named in this Uber driver data breach so far is a driver in the United States. According to SecurityWeek’s November 16 report, the stolen data includes images of driver’s licenses, names, addresses, dates of birth, driver’s license numbers, and other government-issued ID numbers. A filing with the Texas Attorney General’s office lists 149,346 affected people, the largest single figure so far.

Other state filings are smaller and still partial. TechCrunch counted 2,567 people in Massachusetts, 3,430 in South Carolina, and 1,873 in Rhode Island, with notices also filed in Illinois, Iowa, Vermont, and California, where Uber’s filing with the state Attorney General is public. But Uber has not published a total, so treat any national number you see as an estimate until the company states one.

Affected drivers are being offered free credit monitoring and fraud assistance. For the people in those files, the risk is long-lived, since a license image and a date of birth are exactly what an attacker needs to open accounts in someone else’s name.

How did attackers get in?

In the Uber hack, attackers got in with valid credentials. Fireworks AI says an unauthorized party used its internal credentials in early October, and Uber says those credentials let the attacker reach the Uber data stored with the vendor. Alerting at Uber was switched on, yet it stayed quiet, because a login with real credentials looks like the vendor doing its job.

Fireworks has attributed its incidents to a financially motivated group that sells stolen access to large language models, a practice known as LLMjacking. In a November 10 update reported by BleepingComputer, Fireworks also disclosed a second, separate intrusion. A third-party cloud provider hosting its research machines was breached on October 23, and the next day the attacker used that foothold to run an unauthorized workload on Fireworks development machines. That intrusion exposed customer billing details, resource metadata, and, for some customers, datasets and incomplete model weights. Fireworks says payment card numbers were not affected because it does not store them, and it is investigating with Palo Alto Networks’ Unit 42.

No group has publicly claimed the theft. And the exact way the first set of credentials was obtained has not been disclosed.

Was a third party involved?

Yes, at two levels. Fireworks AI was Uber’s third party, holding Uber data to provide AI services. Fireworks, in turn, depended on its own cloud provider, which is a fourth party from Uber’s point of view, and that provider’s breach is what opened the second intrusion. Uber, Notion, and Cursor each lost data through a vendor they chose.

The three victims sent Fireworks very different things. Notion used the platform to run some open-weight models and to analyze stored AI conversations for an admin feature. Notion says AI chat transcripts from fewer than 1% of its users were exposed, even though Fireworks had contractually committed to zero data retention for Notion data. Cursor kept a storage bucket for sharing information with Fireworks, and an access key stored at Fireworks let attackers copy it. Credentials found inside that data were later used to place malicious builds on Cursor’s extension marketplace, a chain we covered in our analysis of the Cursor marketplace attack.

Panorays research traced the chain across the public disclosures from Fireworks AI, Uber, and Cursor, and the shared dependency is cloud inference and model hosting. Companies hand an AI platform storage access keys, datasets, or API secrets so it can run evaluation and inference jobs, and whoever holds the vendor’s credentials inherits that reach. Fireworks’ own updates, reported by The Hacker News, put unauthorized activity on its systems between October 6 and October 25, close to three weeks. In MITRE ATT&CK terms, the way in was valid cloud accounts (T1078.004) followed by collection from cloud storage (T1530), so no malware had to touch the victims’ own networks. Where Fireworks sits for you decides your exposure: it is a third party if you use it directly, and a fourth party if one of your suppliers does.

For risk teams, the lesson sits in the gap between the contract and the platform. A zero-retention clause did not stop retained data from being stolen, and a well-run alerting program did not catch an attacker holding the vendor’s own keys. AI providers now receive identity documents, customer conversations, and engineering data, often through an integration a product team set up quickly, and many of them never pass through a full vendor review. Panorays’ 2025 CISO Survey found that just 3% of organizations had full visibility into their third-, fourth-, and nth-party relationships. Two years later, that visibility gap is even harder to ignore. Uber and Notion are cases in point: both discovered their exposure only after being notified by a vendor, underscoring how easily risks beyond direct suppliers can go undetected.

Mapping that exposure is the work third-party cyber risk management is built for. For every vendor, a risk team needs to know what it has been given, what its credentials can reach, and which providers it depends on. Scoring each relationship by the data and access it carries is what lets a team find the AI vendor that holds driver’s license images before an attacker does.

What should security and risk teams do now?

Start with an inventory of every AI provider that touches your data, then work outward to what each one holds, how its access is guarded, and what your contract lets you verify. None of this needs a new tool or a new team. A small risk team can work through the five steps below in a week, and the first two usually surface vendors nobody had listed.

  1. Find your AI providers, including the ones procurement never saw. Ask engineering, product, and data teams which model hosting, inference, or fine-tuning services receive company or customer data, and add them to your vendor inventory with an owner. The AI vendor risk assessment questions we published earlier give a starting set.
  2. Ask each one what it holds and how its access is guarded. Request a list of the datasets, logs, and conversations it stores and for how long, where the keys that reach your systems are kept and how often they rotate, which cloud and hosting providers sit underneath it, and whether it has checked for exposure to the Fireworks AI incidents.
  3. Treat vendor-held keys as your own attack surface. Rotate any API keys, tokens, and storage credentials you have issued to AI vendors, scope them to the minimum the service needs, and alert on use from unexpected locations or at unusual volumes, since a valid login was enough here.
  4. Check the contract against what the platform actually does. Retention, deletion, breach-notification timelines, and the right to audit subprocessors only help if you can verify them. Ask for evidence of deletion when a project ends.
  5. Rehearse the vendor-breach path. Uber heard about this breach from its vendor and cut the connection the same day, so plan for the call that comes from outside. Our guide to third-party incident response covers who decides, what gets cut off, and when customers or regulators hear about it.

What does the Uber breach reveal about third-party risk?

The Uber incident highlights a challenge that extends far beyond one company: when a third party is breached, the impact doesn’t stop with that supplier. Sensitive information can be exposed across multiple organizations, often without those organizations having a clear picture of their own exposure.

Understanding third-party risk means knowing more than which vendors you work with. It means understanding what data they can access, which other providers they depend on, and how a breach further down the supply chain could affect your business. Without that visibility, organizations may only discover their exposure after an incident has already occurred.

The lesson is simple: you can’t manage the risks you can’t see, especially when they extend beyond your direct suppliers.

For a clearer view of which vendors hold your data and who they depend on, see how Panorays approaches AI and third-party cyber risk management.

Uber Data Breach FAQs

Was Uber hacked?

Uber’s own systems were not reported as breached. Attackers used stolen credentials belonging to Fireworks AI, a vendor Uber used for AI services, to access Uber data stored with that vendor on October 7, 2026. Uber says the credentials did not give access to its other systems, and it cut its data connections with the vendor once it was told on October 9.

What data was leaked in the Uber breach?

The stolen records belong to Uber drivers and include driver’s license images, names, addresses, dates of birth, driver’s license numbers, and other government-issued ID numbers, according to Uber’s state notices as reported by SecurityWeek. Uber is offering affected drivers free credit monitoring and fraud assistance.

How many people were affected by the Uber data breach in 2026?

Uber has not published a total. The largest state figure so far is 149,346 people in a filing with the Texas Attorney General’s office, reported by SecurityWeek on November 16, 2026. Massachusetts, South Carolina, and Rhode Island filings add a few thousand each, and notices were also filed in California, Illinois, Iowa, and Vermont, so the final count may rise.

What is Fireworks AI, and which other companies were affected?

Fireworks AI is a platform developers use to run and fine-tune open-source AI models. It disclosed two intrusions in October 2026, one through stolen internal credentials and one through a breached cloud provider. Besides Uber, Notion confirmed exposed AI chat transcripts, and Cursor said a key stored at Fireworks led to stolen data and malicious extension builds.

Is there an Uber data breach lawsuit or class action?

As of November 16, 2026, we found no Uber data breach lawsuit or class action filed over this incident, and neither Uber nor Fireworks AI has announced a settlement. State notifications are still coming in, so the number of affected drivers may grow. We will update this page if a suit or settlement over the Uber or Fireworks AI breach is reported.

What should companies ask vendors that send data to AI providers?

Ask which AI providers receive your data, what they store and for how long, where the credentials that reach your systems are kept and how they rotate, which cloud providers sit beneath them, and whether they were affected by the Fireworks AI incidents. Ask for evidence, such as deletion records or key-rotation logs.

Subscribe to our blog

You May Be Interested In

The Fastest and Easiest Way
to Do Business
Together, Securely