Uber driver's licence breach: 149,346 notified after Fireworks AI credentials were stolen
Uber says a vendor's stolen internal credentials opened its storage of driver's licence images and details; its alerts never fired because the access looked valid.
· 8 min read
Written by AI from the sources linked below. How we use AI
On this page
Uber is notifying close to 150,000 drivers that images of their driver's licences, and the personal details printed on them, were taken in a breach at a technology vendor. TechCrunch identifies the vendor as Fireworks AI, the company that hosts open-weight AI models for developers and that has already been tied to the Notion and Cursor incidents we covered earlier. The vendor's own compromised credentials were enough to reach a cloud storage area Uber had built for it, and Uber says its alerts never fired because the access looked legitimate. It is a clean example of how a breach at an AI supplier can become a breach of a customer's customers.
Key points
- Uber's notice to the Iowa Attorney General, dated 10 November 2026, says a service provider told Uber on 9 October that an unauthorised party had compromised the provider's internal credentials and used them to reach Uber's account with it.
- Uber's investigation found the credentials were used on 7 October to access Uber data; the data included driver's licence images plus extracted text: name, address, date of birth and licence number.
- TechCrunch reports, citing a filing with the Oregon Attorney General, that 149,346 people are being notified, and names Fireworks AI as the vendor; Uber's own letter to Iowa does not name the provider and counts about 2,999 Iowa residents.
- Uber says the compromised credentials did not give access to other Uber systems, and that it deleted or rotated the credentials tied to the provider. Its investigation is still ongoing.
- Fireworks says its wider incident involved stored secrets, billing metadata and, for some customers, partial model weights and datasets, and links it preliminarily to a financially motivated group selling access to AI models.
What to do
If you drive for Uber and received a letter: the letter comes from Uber through a mailing vendor (Cyberscout) and begins on 10 November. Read what it offers, and treat the exposed items as permanently compromised: a name, address, date of birth and licence number are enough to attempt identity fraud in your name. Consider a credit freeze with the credit bureaus in your country, and be suspicious of calls, texts or emails that quote your licence number or claim to be from Uber, a licensing agency or your bank.
If your company is a Fireworks customer: Fireworks says it is notifying impacted customers directly and that recommended actions will be in that email. Its bulletin gives security-notice@fireworks.ai for questions. Independently of that, rotate any API keys, evaluation-job secrets and cloud credentials that were ever stored with Fireworks, and review who and what can reach storage you created for a vendor.
If you run vendor integrations: check whether vendor-owned credentials can read your buckets, and whether your access alerts can tell a legitimate vendor session from a stolen one. Uber's alerts could not.
What happened
Uber's letter to the Iowa Attorney General, prepared by outside counsel and dated 10 November, sets out the sequence. On 9 October a service provider told Uber that an unauthorised third party had compromised the provider's internal credentials and used them to access Uber's account with the provider. Uber says it immediately severed data connections with the provider and opened an investigation. On 13 October the investigation concluded that the intruder had used the compromised credentials on 7 October to access Uber data containing personal information connected to the provider.
The data was driver identity material: images of driver's licences and text extracted from them, including name, address, date of birth and licence number. According to the letter, it sat in a cloud storage environment Uber created specifically to let the provider do its work. That environment was covered by Uber's cloud security management processes and watched by its cloud security team, and it had alerting turned on. But because the intruder used valid credentials and data flows that the provider's own compromised system had enabled, the letter says no alerts were triggered. Uber adds that the credentials did not give access to other Uber systems.
After the incident Uber deleted or rotated credentials and access information associated with the provider, and the provider told Uber it had revoked the compromised credentials. Individual notices began going out on 10 November.
TechCrunch's reporting adds the vendor's name and the scale. It says Uber's statement referred to "a security incident impacting one of our third-party vendors, Fireworks.ai", and that a filing with the Oregon Attorney General puts the number of people notified at 149,346. TechCrunch also says Uber has not given an overall total and that it is unclear whether the number will grow. TechCrunch lists drivers notified in California, Illinois, Massachusetts, Rhode Island and South Carolina, with further filings in Texas and Vermont. BleepingComputer's headline, which we could not open in full, also puts the figure at nearly 150,000 people.
The numbers fit together: the Iowa letter counts 2,999 Iowa residents, a small slice of a nationwide total near 150,000 spread across many states. We have not seen the Oregon filing ourselves, so the 149,346 figure rests on TechCrunch's account of it.
Background
Fireworks AI sells fast inference and fine-tuning for open-weight models, which makes it attractive both to developers and to criminals who want cheap access to models. The company published its first security bulletin on 9 October, the same day Uber says it was alerted. In a bulletin that it last updated on 9 November and still lists as under active monitoring, Fireworks describes two phases. On 23 October a third-party cloud provider hosting Fireworks research and development machines had a security incident. On 24 October the unauthorised party used that access to reach Fireworks development machines and run an unauthorised workload that caused additional customer data impact, which Fireworks detected the same day.
The dates matter. The access to Uber's data happened on 7 October, before the cloud-provider incident Fireworks dates to 23 October, so Uber's exposure belongs to the earlier part of a longer intrusion. Fireworks has not said how the first compromise happened, and Uber's letter says only that the provider's internal credentials were compromised.
The wider fallout has been tracked in our earlier coverage. We reported on the Fireworks breach and Notion's statement that under 1% of workspaces were hit, and on how a Fireworks key led to a trojanized Claude Code release through the Cursor marketplace. TechCrunch says Notion and Cursor have both confirmed data was stolen in the Fireworks hack, and Uber's notices add a third well-known customer to that list.
How the access worked
The mechanics are what make the case instructive. Uber did not hand Fireworks a copy of its data on Fireworks' servers; it created a storage environment on its own side and gave the vendor a route into it. That is a common and sensible design, because the data stays under the customer's security tooling. The weakness is that the vendor's identity becomes a key to the customer's door.
When that identity was stolen, the intruder's requests looked like the vendor's normal work. Uber's letter explains that the alerting was enabled but did not trigger because the third party used valid credentials and the data flows the compromise of the provider's system had enabled. Detection rules built to catch outsiders or unusual locations do not catch an insider's key used in a familiar way. Detecting this class of attack generally needs rules about behaviour: sudden bulk reads, unusual volumes of identity images leaving, or access outside the vendor's normal schedule. Whether Uber had such rules is not stated in the letter.
Fireworks says its stored secrets were among the first items known to be accessed, specifically secrets used in evaluation jobs, and later extended to a limited set of accounts. It also lists resource metadata such as names and creation times, and billing information including billing and shipping addresses, invoice and payment history and credit usage. It says it does not store payment card numbers and that they were not affected. For certain customers a subset of incomplete model weights and datasets was accessed. It says its inference, Training SDK and Nexus services, which are zero data retention by default, were not impacted according to current findings. It has disabled self-serve dataset uploads and downloads for batch inference and dataset-based training while it investigates.
What it means
For drivers, the practical problem is that driver's licence data is hard to change. A password can be rotated; a date of birth and licence number cannot. The licence images are more sensitive than the text because they show the document itself. The harm depends on whether the data is sold or used, and the notices we saw do not report misuse.
For companies, the case adds weight to a familiar lesson: the security of your data is the security of your weakest vendor with a credential. Teams that adopted AI vendors quickly in the past two years should now check what those vendors can read. Questions worth asking: Does the vendor hold standing credentials, or short-lived tokens? Is our data copied to them or accessible in place? Do our logs let us tell the vendor's real sessions from a thief's? How quickly will they tell us, and how quickly can we cut the connection? Uber's timeline is a decent reference: severed on 9 October, scoped by 13 October, notified by 10 November, a month between the finding and the letters.
For AI infrastructure providers, the incident shows that inference hosts collect more than model traffic: they hold customer secrets, billing records, datasets and weights, which makes them targets for groups after model access. Fireworks says the evidence indicates links to a financially motivated group that aims to sell tokens from unauthorised access to AI models to illicit routing gateways, a practice it calls LLMjacking. That attribution is preliminary and the investigation is ongoing.
What we don't know yet
- Whether the 149,346 figure will rise, and how many of those people are in each state beyond the ones reported.
- Why Uber's drivers' licence images were held in a vendor-accessible store, and what work the vendor was doing with them. Uber's letter says only that the environment was created to enable the provider's work.
- How the provider's internal credentials were first compromised, and how the 7 October access relates to Fireworks' 23 and 24 October timeline.
- Whether any of the stolen data has been used or sold. No group has claimed it, according to TechCrunch.
- Whether regulators or courts will act. We have found no such action as of this writing.
What to watch next
Watch Fireworks' bulletin, which it keeps updating, for the end of active monitoring and for a count of affected customers. Watch the state attorney-general breach databases for further Uber filings and for new Fireworks customers disclosing their own exposure. And watch whether Uber, or other customers, change how vendors are given access, for example by moving to short-lived credentials.
Sources
- Uber's notification letter to the Iowa Attorney General, 10 November 2026 (PDF)
- Fireworks AI security bulletin, October 2026 incident update
- TechCrunch: Uber says hackers stole information on close to 150,000 driver's licenses
- BleepingComputer: Uber notifies drivers of data breach affecting nearly 150,000 people
- audiotale.net: Fireworks AI breach
Written by
The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.
Read next
Ubuntu Pro Edge archive ships AI-written CVE fixes from Canonical's Surge
Canonical's opt-in Edge archive gives Ubuntu Pro users patches from its multi-agent Surge framework before full testing; stable security updates stay human-validated.
audiotale.net team6 min read
Anthropic cuts bug disclosure window to 30 days for high and critical flaws
From 1 January 2027, Anthropic will publish high and critical vulnerabilities its Claude models find in open-source software after 30 days instead of 90, citing faster fixes and bug rediscovery.
audiotale.net team7 min read
Google GTIG: exploit_agent turns patches into exploits, no human involved
Google's threat team says attackers ran an AI framework that diffs patches, writes exploits and runs them automatically, the first fully autonomous pipeline it has seen in the wild.
audiotale.net team8 min read