Skip to content
All posts

Uber Breach via Fireworks AI: 149,346 Notified, What to Do

November 18, 2026 · Team PlayCISO
TL;DR · the short version

Uber Technologies is notifying 149,346 people, per its Oregon attorney general filing as reported by TechCrunch, after attackers used credentials stolen from its AI provider Fireworks AI to reach Uber data held there. Uber's notice letter (10 November 2026) says the provider told Uber on 9 October that its "internal credentials" had been compromised and used to access Uber's account; Uber cut its data connections and on 13 October found that Uber data was accessed. TechCrunch, citing state filings, reports names, addresses, dates of birth and driver's license numbers. Uber had alerts on its Fireworks environment, but none fired because the attackers used Fireworks' own credentials. Fireworks says stolen staff credentials and a flaw in one of its services gave access between 3 August and 7 October, with environment variables (API keys and tokens) captured in some customer accounts, and that a second, separate intrusion from 23 to 30 October exposed account and billing data for all accounts created before 30 October. Uber offers 90-day enrollment in Cyberscout credit monitoring and says it has seen no misuse so far. Notion and Cursor also confirmed data access.

Put it into practice
Plan the rotation before the next AI vendor notice
Map what each AI provider key touches, who owns it and the order to replace it, so a vendor bulletin is a routine change.
Open the Key Rotation Planner

Uber Technologies is notifying people that their personal information, including driver's license details, was accessed after an unauthorized party used Fireworks AI's own internal credentials to reach Uber's account with that AI provider. Uber says the provider told it on 9 October 2026; Uber confirmed on 13 October that Uber data was accessed and began mailing letters dated 10 November. Uber's filing with the Oregon attorney general lists 149,346 people, per TechCrunch. Fireworks has confirmed unauthorized access affecting customer data; no group has claimed the theft.

Independent analysis. PlayCISO is not affiliated with Uber or Fireworks AI. Facts as published on 18 November 2026; both companies say their investigations continue.

Confirmed vs not yet known

Confirmed by Uber (notice letter): a service provider's internal credentials were compromised and used to access Uber's account with the provider; Uber was told on 9 October 2026, cut its data connections, and on 13 October found that Uber data connected to the provider was accessed. Free Cyberscout credit monitoring, 90 days to enroll. No misuse "to date".

Confirmed by an Uber spokesperson (to TechCrunch): the provider is Fireworks.ai, "which had access to some Uber data".

Confirmed by Fireworks (security bulletins): stolen Fireworks credentials and a vulnerability in one of its services were used between 3 August and 7 October; environment variables, including API keys and tokens stored as secrets, were captured in some customer accounts; a second, separate intrusion (23 to 30 October) reached account and billing data for all accounts created before 30 October, and datasets, models or secrets for a small number of customers.

Reported from state filings, not in the published letter: the 149,346 total (Oregon) and the data elements: names, addresses, dates of birth, driver's license numbers (TechCrunch), plus license and ID images (Security Affairs).

Not yet known: which of Fireworks' two incidents reached Uber (the 7 October breach date fits the first, but neither company says so); why driver's license data sat with an AI model provider and in what form (a dataset, a fine-tuning job, evaluation data); how many of the 149,346 are drivers; who did it.

What happened

Uber's letter is short and does not name the provider. It says that on 9 October 2026 "a service provider notified Uber that an unauthorized third party had compromised the service provider's internal credentials and used them to access Uber's account with the service provider." Uber "immediately severed data connections with the service provider", and on 13 October its investigation "determined that the unauthorized party had used the compromised credentials to access Uber data connected to the service provider." The California filing gives the breach date as 7 October 2026.

An Uber spokesperson told TechCrunch: "The notices to drivers are in relation to a security incident impacting one of our third-party vendors, Fireworks.ai, which had access to some Uber data." TechCrunch also reports that Uber had alerts enabled to detect unauthorized access to its Fireworks environment, but "no alerts were triggered" because the attackers used Fireworks' internal credentials.

That is the lesson of this incident. The credential that failed was not Uber's API key; it was the provider's own. The customer's detection was built around the customer's identities, and the attacker used the provider's.

Useful next, if this was relevant to you

Showing 1 of 3: Breach Cost Calculator

What Fireworks AI says

Fireworks keeps one bulletins page with three entries:

  • 10 October: on 6 October Fireworks identified "unauthorized use of internal Fireworks credentials". An unauthorized third party accessed environment variables used in evaluation jobs in a limited number of customer accounts; Fireworks believes "environment variables, including API keys and tokens stored as secrets, were captured". It rotated employee API keys on 9 October and revoked more user API keys and service-account tokens on 10 October.
  • 9 November: the impact was wider: account and billing information and, for a limited set of accounts, secrets, datasets and incomplete models. Fireworks links the activity to a financially motivated group selling tokens from unauthorized access to large language models to "illicit routing gateways", known as LLMjacking. It brought in Palo Alto Networks Unit 42 and notified law enforcement. It also says it has not detected impact to its Inference, Training SDK and Nexus products, which it describes as zero data retention by default.
  • 17 November: two separate incidents. The first (3 August to 7 October) used stolen Fireworks credentials and a vulnerability in one of its services; for a small number of customers, secrets, model files and datasets were accessed. The second (23 to 30 October) came through Fireworks development machines hosted at a third-party cloud provider that had its own incident, reached part of production, and accessed account and billing information for all accounts created before 30 October. Fireworks says it does not store payment card numbers. Both incidents are described as contained and remediated.

Timeline

Date (2026)EventSource
3 Aug to 7 OctFirst Fireworks incident: stolen credentials plus a service vulnerabilityFireworks bulletin, 17 Nov
6 OctFireworks identifies unauthorized use of internal credentials, starts containmentFireworks bulletin, 10 Oct
7 OctBreach date in Uber's filingCalifornia AG listing
9 OctFireworks notifies Uber; Uber cuts its data connections. Fireworks rotates employee API keys and publishes its first bulletinUber letter; Fireworks bulletin
13 OctUber determines Uber data was accessedUber letter
23 to 30 OctSecond, separate Fireworks incident via a third-party cloud providerFireworks bulletin, 17 Nov
9 NovFireworks widens scope: billing data, datasets, incomplete models; LLMjacking linkFireworks bulletin, 9 Nov
10 NovDate on Uber's notice lettersUber letter (California and South Carolina copies)
13 NovFirst reports from state filings; Uber names FireworksTechCrunch
16 NovOregon filing shows 149,346 people notifiedTechCrunch
17 NovFireworks says the activity was two separate incidentsFireworks bulletin, 17 Nov

What security leaders should do this week

  • List every AI provider that holds your data or your secrets. Inference APIs, fine-tuning platforms, evaluation services and model routers all count. For each, write down what is stored at rest: datasets, uploaded files, evaluation jobs, environment variables. The AI supplier questionnaire gives you the questions.
  • Rotate anything stored with an affected provider. If you used Fireworks evaluation jobs or stored secrets there, rotate those keys and tokens now, whether or not you received a notice, and check provider usage for spend you do not recognise. Plan the order with the Key Rotation Planner.
  • Scope the keys you hand to AI vendors. A key placed in a provider's environment variables should reach only what the job needs, expire quickly, and be one you can revoke without breaking production.
  • Scan for keys in the places AI work happens. Notebooks, evaluation configs, prompt files and CI logs leak keys more often than application code. Run the Secret Scanner over them, and see TokenPwn: check if your LLM API key was leaked for the LLMjacking angle.
  • Restrict egress from the provider side. Where a provider connects back into your data (storage buckets, warehouses, connectors), allow-list its addresses and scope the connection to the data the job needs. Uber's first move was to cut those connections; know how you would do the same in minutes.
  • Stop sending identity documents to AI vendors unless you must. Find out whether driver's licenses, ID images or other regulated data are in any dataset or evaluation set sent to a model provider, and minimise or tokenise it.
  • Ask how the provider watches its own staff. Your alerts did not fire for Uber. Ask each AI provider how it detects misuse of its employees' and service accounts' access to your tenant, and whether it will give you those logs.
  • Check the contract. Confirm breach-notice deadlines, what the vendor must tell you (data, dates, indicators), help with your own regulator and customer notices, and who pays for credit monitoring. Run the vendor side with the third-party breach response checklist.

Why it matters beyond Uber

AI providers have become a new kind of hub. One provider holds keys, datasets and model files for many customers, so one stolen staff credential reaches all of them: Uber, Notion and Cursor in this case, per TechCrunch. It is the same hub-and-spoke pattern described in Black October: lessons from the downstream breaches, now applied to AI. The attackers Fireworks describes want model access to resell; the customer data was taken along the way.

For a wider view of this week's incidents, see Cyber incidents, 17-18 November 2026: what is confirmed, and the related Infoblox breach through a SharePoint migration vendor, another case of a trusted vendor's access being used against its customer.

FAQ

Was Uber hacked?

Not directly, according to Uber. Its notice letter says a service provider told Uber on 9 October 2026 that an unauthorized third party "had compromised the service provider's internal credentials and used them to access Uber's account with the service provider." An Uber spokesperson told TechCrunch the provider is Fireworks.ai. Uber's own systems are not described as breached; Uber data held at the provider was accessed.

What data was taken in the Uber Fireworks AI breach?

TechCrunch, citing Uber's state filings, reports names, home addresses, dates of birth and driver's license numbers. Security Affairs reports the filings also list images of driver's licenses and other government IDs. Uber's published letter template leaves the data elements as a placeholder, so each person's letter says what applied to them.

How many people were affected by the Uber data breach?

Uber's filing with the Oregon attorney general lists 149,346 people in total, per TechCrunch. Uber has not said how many of them are drivers or whether the number will rise. Earlier reporting of the first state filings (Massachusetts 2,567, South Carolina 3,430, Rhode Island 1,873) gave a much lower partial count of 7,870; the Oregon figure replaces it.

Which credentials were compromised: Uber's or Fireworks AI's?

Fireworks' own. Fireworks says an unauthorized party used stolen Fireworks credentials, and its 10 October bulletin says environment variables, "including API keys and tokens stored as secrets", were captured in some customer accounts. TechCrunch reports Uber had alerts on its Fireworks environment but "no alerts were triggered" because the attackers used Fireworks' internal credentials.

What is Fireworks AI and who else was affected?

Fireworks AI runs a platform for using and fine-tuning open-weight AI models. TechCrunch reports Notion and Cursor also confirmed that some of their data was accessed. Fireworks' 17 November bulletin says it was two separate incidents and that, in the second, account and billing information for all accounts created before 30 October was accessed (payment card numbers are not stored).

Who was behind the Fireworks AI breach?

Not known. Fireworks says the evidence links the activity to a financially motivated group that sells tokens from unauthorized access to large language models to illicit routing gateways, known as LLMjacking. TechCrunch reports no known group has claimed the Uber data theft.

What should Uber drivers do?

Enroll in the free Cyberscout credit monitoring within 90 days of the letter's date using the activation code in the letter, review credit reports, and consider a credit freeze. Uber's letter says it has "not identified any misuse" of the data so far. Treat calls, texts or emails about the breach that ask for codes or payment as likely scams.

What should companies that use AI providers do after the Fireworks breach?

List which AI providers hold your data or your secrets, rotate any key or token stored with an affected provider, and ask each provider what it stores at rest (datasets, evaluation secrets, model files) and how it would detect use of its own staff credentials against your account. Uber's experience shows your own alerts may not fire when the provider's credentials are used.

Sources

This article is educational guidance for security leaders, not legal advice. Check Uber's letter and Fireworks' bulletins for the latest details before acting.

Plan the rotation before the next AI vendor notice
Map what each AI provider key touches, who owns it and the order to replace it, so a vendor bulletin is a routine change.
Open the Key Rotation Planner
Take this with you
Confirmed vs not yet known · What happened · What Fireworks AI says · Timeline · What security leaders should do this week · Why it matters beyond Uber

Ready to practise the decisions these articles describe?

Run your first Cyber War Room free →

Pro