Uber Didn't Get Hacked. The AI Middleman Did.
One breach at a $17.5 billion AI inference vendor exposed 150,000 Uber driver's licenses, Notion AI chats, and Cursor's coding traces — and broke the zero-data-retention promise the entire enterprise AI stack is built on.
Over the past week, Uber has been telling close to 150,000 drivers that images of their driver's licenses — along with their names, addresses, dates of birth, and license numbers — had been stolen. Uber's own systems were never breached. No Uber password failed, no Uber server was compromised, and the company's monitoring saw nothing wrong, because from the network's point of view nothing was wrong: the attacker walked in through a data pipe Uber had built, on purpose, for one of its AI vendors.
That vendor was Fireworks AI. Within the same ten days, Notion told customers that transcripts of their Notion AI chats had been exposed through the same company, and Cursor — the AI coding tool — disclosed that an access key it had issued to Fireworks was used to copy a storage bucket full of coding-agent session traces, some containing live customer credentials. Three companies, three separate disclosures, one vendor none of their users had ever heard of.
This is the AI economy's first full-scale supply-chain breach, and it hit the layer of the stack that almost nobody diligences.
The Middleman You've Never Heard Of
Fireworks AI is an inference provider: it runs AI models on rented GPUs so that other companies don't have to. When Notion classifies how your team uses its AI features, when an app serves you an open-weight model's output, there is a reasonable chance the actual computation happened on infrastructure like Fireworks'. The company said in July that it had passed $1 billion in annualized revenue run rate and served more than 40 trillion tokens — claims made while announcing a $1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV, with Nvidia among the participants.
That is the structural fact that matters: a company most consumers have never interacted with became a load-bearing wall under Uber, Notion, Cursor, and — as Notion's own advisory describes it — a platform "powering tens of thousands of customers." Every one of those customers routes data into it. Very few of their security teams treat it as what it has quietly become: a concentration point with the blast radius of a cloud provider and the security track record of a venture-stage startup.
Two Break-Ins in Twelve Weeks
According to Fireworks' own security bulletins, what looked like one incident was actually two.
The first ran from early August to October 7. An attacker used stolen Fireworks credentials plus a vulnerability in one of its services to misuse the platform and, for "a small number of customers," access secrets, model files, and datasets. Palo Alto Networks' Unit 42 ran the forensics; Fireworks says that investigation is complete.
The second ran October 23–30. A third-party cloud provider hosting Fireworks' development machines suffered its own security incident, and the attacker used those machines "to reach part of our production systems." The result: account and billing information for every Fireworks account created before October 30, and — again for a small number of customers — datasets, models, and secrets. A second forensics firm is still investigating.
The motive, per Fireworks, is a financially motivated criminal group engaged in what Google's threat intelligence team calls "LLMJacking" — stealing credentials for large language model access and reselling the tokens through illicit routing gateways. The attackers weren't after Uber drivers' licenses specifically. They were after compute. The personal data was collateral inventory.
Briefings like this are free to read with a free AlphaBriefing membership. Join free →
Three Customers, Three Different Wounds
What makes this breach a case study is that each victim lost something different, because each had wired the vendor into a different part of its anatomy.
Uber lost identity documents. Its driver's license data sat in a cloud storage environment Uber had created specifically for the vendor's work. Uber's notice to state regulators is blunt about why nobody noticed: "The environment had alerting enabled, but because the unauthorized third party used valid credentials and data flows enabled by the compromise on the service provider's system, no alerts were triggered." Oregon's breach registry puts the count at 149,346 people.
Notion lost conversations. The exposed dataset included Notion AI chat transcripts used for its Insights feature, affecting fewer than 1% of workspaces. Notion notes most transcripts were "unattributable" — not labeled with user or workspace. But Notion's own advice to affected customers gives the game away: if you pasted passwords or API keys into Notion AI during the window, rotate them.
Cursor lost the keys to its own supply chain. An access key Cursor had issued to Fireworks — "exposed by Fireworks," in Cursor's words — was used to copy a bucket of coding-agent traces, mostly from 2025. Some contained unredacted API credentials, which the attacker used to reach SaaS tools Cursor uses. Separately, on October 29, malicious versions of the Anthropic Claude Code extension appeared in Cursor's own marketplace for roughly eleven hours, beaconing out to an attacker-controlled server. A breach of an AI vendor became, within days, an attempted backdoor into developers' machines.
The Contract Said Zero Data Retention
Here is the detail that should concern investors more than any single stolen license: Notion had a contractual commitment from Fireworks to zero data retention. Fireworks' marketing says its platform doesn't store prompts or outputs by default. Notion's post-incident finding, in its own words: "during our investigation we learned that data was retained from the Fireworks endpoints that we used."
A contract said the data wouldn't exist. The breach proved it did. Every enterprise that has signed a ZDR clause with an AI vendor — which is to say, essentially every enterprise deploying AI right now — just watched that clause fail in production. Notion is now auditing its critical vendors to verify data-minimization claims rather than accepting them. It will not be the last.
The question for markets is what happens when thousands of procurement departments reach the same conclusion in the same quarter — and who gets paid, and who gets repriced, when "trust but verify" becomes a budget line.
A free membership gets you access to every free article at AlphaBriefing.com: Join AlphaBriefing
The rest of this briefing is free — it just requires a free AlphaBriefing account: the earnings print that shows where the security budget is already flowing, what this does to the private AI-infrastructure trade and the $17.5 billion question mark now hanging over it, and the five-item watchlist for how this cascades through December.
Create your free account → — 30 seconds, no card.