Thousands of Uber drivers had licence records stolen after hackers broke into AI supplier Fireworks AI, and Cursor users were handed a booby-trapped coding tool in the intrusion. Both companies disclosed the damage this week.
Key Points
- Hackers stole driver’s licence records for at least 7,870 Uber drivers across Massachusetts, South Carolina and Rhode Island, with more state filings still to come and no nationwide total disclosed.
- The same intrusion at Fireworks AI gave attackers a key that Cursor had issued to the vendor, which they used to slip malicious copies of Anthropic’s Claude Code extension into Cursor’s marketplace for about 11 hours on 29 October.
- At most 5,000 Cursor users may have received the tampered extension, versions 2.1.312 and 2.1.313, before it was pulled. Anthropic did not publish the copies and its product was not at fault.
- Notion has also confirmed that its data was taken in the Fireworks intrusion, affecting fewer than 1 percent of its workspaces.
- Fireworks says the attackers were after tokens that unlock paid access to large language models. No group has claimed responsibility, and Uber says it has seen no misuse of the stolen driver data so far.
Thousands of Uber drivers had licence records stolen after hackers broke into AI supplier Fireworks AI, and Cursor users were handed a booby-trapped coding tool in the intrusion. Both companies disclosed the damage this week.
One vendor, three victims, and a single point of failure. The picture that emerged on 13 November ties together a ride-hailing giant, a popular coding tool and a workplace app, all through the same supplier. It is the latest warning that the companies building and running AI systems now hold the keys to far more than their own data.
What happened
Uber learned of the trouble on 9 October, when its provider told the company that hackers had taken internal login credentials and used them to reach Uber’s data. By 13 October, Uber’s investigators had confirmed that personal information had been accessed. Letters to affected drivers, dated 10 November, went out this week, and filings with several state attorneys general, including California, put the first numbers on the damage.
The filings seen by TechCrunch show driver’s licence details taken for 2,567 people in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island, a total of at least 7,870. An Illinois filing shows some residents there were also affected, but gives no number, and Uber has not said how many drivers are affected across the country. The California notice does not name the provider or list the data fields involved. Uber cut its data links to the vendor as soon as it learned of the intrusion, and it is offering affected drivers free credit monitoring and fraud help through Cyberscout, a TransUnion company.
TechCrunch identified the vendor as Fireworks AI, a company that gives developers access to open AI models, and an Uber spokesperson confirmed it. A Fireworks spokesperson confirmed that unauthorised access to its systems had affected customer data. Fireworks says the intruders were hunting for tokens that buy access to large language models, a theft known as LLMJacking, and Palo Alto Networks’ Unit 42 is helping it trace the intrusion.
Cursor’s account, published on 13 November after a two-week investigation, shows the same break-in reaching further. An attacker obtained an access key Cursor had issued to Fireworks, and Cursor says the key was exposed by Fireworks. The key opened a read-only storage bucket Cursor used to share data with the vendor. It held limited coding-agent traces, mostly from 2025, from sessions where users had not switched on Privacy Mode. A small number of those traces still contained credentials that had not been removed, and fewer than 0.01 percent of customers had API credentials exposed.
From there, the attacker reached some of Cursor’s software tools and then, on 29 October, placed malicious copies of Anthropic’s Claude Code extension, versions 2.1.312 and 2.1.313, directly on the cache that serves Cursor’s VS Code extension marketplace. At most 5,000 users who asked for those cached versions may have received the tampered build during the roughly 11 hours before Cursor removed it. The company took the marketplace offline, rebuilt it, and worked with the cloud provider hosting the attacker’s server to take that down. Cursor stresses that Anthropic did not publish the copies, that this was not a flaw in Anthropic’s product, and that the marketplace source, Open VSX, never carried the malware. The Claude tools at the centre of the incident are now spreading into factory floors as well as software teams, which raises the stakes of any tampering.
Notion, the workplace software company, has also confirmed that its data was taken in the Fireworks intrusion, affecting fewer than 1 percent of its workspaces. Fireworks first disclosed its own breach in October, and its security bulletin, last updated on 9 November, does not mention Cursor.
Why it matters
The damage came not through Uber’s own systems, nor through Cursor’s, but through a supplier both trusted with access. That is the pattern security teams fear most: a small vendor with broad permissions becomes the way in, and the breach spreads to every customer that handed it keys. In this case the same intrusion produced stolen identity documents on one side and a poisoned developer tool on the other, two very different kinds of harm from one weak link.
The motive adds a new twist. The attackers were not after Uber drivers or Cursor users in particular. Fireworks says they wanted tokens that unlock paid AI models, so they could sell access to them. The driver licences and the coding traces were, in effect, collateral: data sitting in systems the attackers passed through on the way to something they could sell. As AI suppliers hold more customer data and more credentials, that kind of opportunistic theft gets more valuable, and the blast radius of a single vendor breach gets wider.
For developers, the Cursor episode is the sharper warning. A tampered extension can run code on a machine with the same trust a user gives their editor, and the malicious build sat in the marketplace cache for 11 hours. Cursor says it has contacted every affected customer directly, moved all of its inference traffic off Fireworks, deactivated the key, and rotated credentials across its systems. It also says the last attacker activity it saw was on 1 November, and that it was blocked.
The timing matters for the AI industry too. The same week, investors were piling into the sector, with orders for Anthropic’s stock market debut reaching about six times the shares on offer. Growth and trust are now moving in opposite directions: money is flooding in while the supply chain underneath it shows cracks. Companies that sell safety testing, such as OpenAI’s new Daybreak red-team unit, are finding no shortage of work.
What happens next
Uber’s driver notices are still being filed state by state, and those filings will give the first full count of how many drivers were exposed. Until then, the 7,870 figure covers only three states, and the real total is likely higher. Affected drivers have 90 days from their letter to sign up for the free credit monitoring Uber is offering.
Cursor users who installed or updated the Claude Code extension from its marketplace on 29 October should follow the steps in the company’s 30 October security notice. The company warns that simply removing the extension may not be enough. It says it will contact customers directly if it finds anything further that affects them, and it has brought in outside forensics experts to support the investigation.
Fireworks has not said how many customers were affected in total, and no group has claimed the attack. The episode is now a test case for how AI vendors handle the credentials customers hand them, and for whether the industry tightens those arrangements before the next supplier becomes the way in.
Sources: TechCrunch, Reuters, Cursor, Gizmodo. 14 November 2026.



