What to know: Uber says the incident involved Fireworks AI, and a state filing reported by TechCrunch puts the number of people notified at 149,346. If you receive a notice, verify it through the company’s official app or website and follow its instructions. Exposure does not by itself mean your identity has been misused.
Uber Driver Data Stolen in Fireworks AI Breach as Notion Reveals Exposure
Nearly 150,000 people were notified after driver information was stolen through an AI supplier. Fireworks says it has since identified two separate security incidents, with other customers affected in different ways.

Jump to a section
A driver can hand a license to Uber without ever choosing an AI company. Yet information from that license can end up in the hands of a supplier working behind the scenes. When the supplier’s credentials were compromised, that relationship became part of the risk. This kind of supply-chain attack reaches people through a service they depend on.
Uber told TechCrunch that the incident involved Fireworks AI, a provider of AI infrastructure. The publication reported on November 16 that a notice filed in Oregon put the number of people notified at 149,346. The information stolen included names, addresses, dates of birth and driver’s license numbers. Affected people should check their individual notices for what applies to them.
On November 17, Fireworks said its investigation had identified two separate incidents, both of which it considers contained and remediated. That distinction matters: they had different routes into the company, and the information affected depended on the customer and incident.
How one supplier became a route to sensitive data
Uber’s notice, as reported by ITPro, says compromised provider credentials were used to access an Uber cloud environment created for the supplier’s work. The affected information included images of driver’s licenses and extracted details. Uber said the use of valid credentials and permitted data flows did not trigger its alerts. The compromised credentials did not provide access to its other systems, according to the notice.
SecurityWeek’s account of Uber’s notices gives three useful dates: the access to Uber data occurred on October 7; the provider alerted Uber on October 9; and Uber’s investigation determined on October 13 that its data had been accessed. Uber then notified affected people in November. Fireworks’ second incident began later, on October 23. The public notices do not establish that the October 7 access occurred during that later incident.
This is the uncomfortable distinction between a valid login and an authorized person. A system may recognize the key and the path while failing to recognize that the person using them has changed. That does not mean monitoring is useless. It means security teams also need to ask what a vendor account can reach, whether it still needs that access and how unusual use would be detected.
Fireworks says the first incident ran from August 3 to October 7 and involved stolen credentials and a vulnerability in one of its services. For a small number of customers, the attacker accessed secrets, model files and datasets. In a separate incident beginning October 23, Fireworks says an intruder reached its development machines through a breached cloud provider and then part of its production systems. Fireworks says account and billing information for accounts created before October 30 was accessed in the second incident, along with datasets, models or secrets for a small number of customers. It says it does not store payment-card numbers.
The public descriptions do not establish that an AI agent carried out either intrusion. Nor should a report about one customer’s data be treated as proof that every Fireworks customer lost the same information.
Notion’s separate warning: what people told an AI tool
Notion says the Fireworks incident affected fewer than 1% of its workspaces. The exposed dataset included Notion AI chat transcripts used for its Insights feature. For the vast majority of affected workspaces, the transcripts were not labeled with the author or workspace, although the text itself could contain identifying information.
Notion says it contacted affected customers and recommends treating the identified data as exposed. It says affected customers can review the relevant data securely in their workspace. If a person pasted passwords, API keys or other credentials into affected conversations, the company recommends rotating them. People who have not received a notice have not been identified as affected based on its investigation, which remains ongoing.
The two examples are related through a supplier, not identical in their consequences. An exposed license can create an identity-fraud concern; an exposed work chat can reveal sensitive business information or a credential. The right response depends on the data actually involved.
The AI connection is about access, too
Fireworks previously said evidence linked the activity to financially motivated actors seeking to sell unauthorized access to large language models, a practice known as LLMjacking. Google’s Mandiant team has described separate investigations in which stolen cloud identities were used to consume or resell AI capacity. That research explains the wider technique; it does not prove the same operators or every detail of the Fireworks incidents.
Our earlier reporting on stolen AI keys examined how access can be abused. The Fireworks case adds another question: what else can the same trusted relationship reach? Our Optimizely coverage describes a different supplier incident and a different mechanism. Neither report establishes a connection between the attackers.
What affected people and businesses can do
A data breach is an incident in which sensitive information is accessed or exposed without authorization. The response depends on the type of information and the evidence of misuse.
- Verify the notice independently. Open the service’s official app or type its website yourself. Do not share login codes or identity documents with someone who contacts you unexpectedly about the breach.
- Read what was exposed. A driver’s license, a password and an AI conversation create different risks. A breach notice is not proof that an account was taken over or that money was stolen.
- Consider identity protections. In the United States, the FTC explains free credit freezes and fraud alerts. A freeze can make it harder to open new credit in your name, but it will not stop every form of identity fraud. Ask your state motor vehicle agency about its process if your license information was exposed.
- If credentials appeared in an affected chat, revoke or rotate them. Use a trusted device, follow the issuing provider’s instructions and review relevant sessions or access logs. For a work account, involve the organization’s security team. Follow Notion’s notice to review the exposed data in your workspace.
- Map supplier access. Organizations should inventory connected vendors, restrict credentials to the access needed, remove unused access and review activity. Preserve relevant records and contain any active compromise promptly.
Our data breach response guide explains how to check a notice and prioritize next steps. For businesses, our incident response guide covers containment and recovery.
AI suppliers can be several steps removed from the people whose information passes through a service. That distance makes it more important, not less, to understand the access a supplier receives.