A stolen Fireworks AI key led to a fake Claude Code extension in Cursor's marketplace
In an incident report, Cursor confirmed that an access key exposed on the Fireworks AI side was stolen, and that the attacker used the chain to place malicious copies of the Claude Code extension in its marketplace on October 29, 2026. The builds were up for about 11 hours, and Cursor estimates at most 5,000 users may have received them.
A leak at a supplier reached the code editor's extension store
Cursor, the AI-assisted code editor, named the source of the attack in an incident update it published on Friday. According to Cursor, an attacker got hold of an access key the company had issued to Fireworks AI, a provider it used for model training and inference. The key had been exposed on Fireworks' side.
Using that key, the attacker copied data out of a read-only bucket that Cursor shared with Fireworks. Credentials found in the data opened some internal systems, and on October 29, 2026 the attacker placed tampered copies of Anthropic's Claude Code extension in the cache that serves Cursor's VS Code extension marketplace. BleepingComputer reported it on November 14.
The versions, the scope and the exposed data
The rogue builds were numbered 2.1.312 and 2.1.313 and, per Cursor, stayed reachable for roughly 11 hours. Cursor says at most 5,000 users who requested particular cached versions on October 29 may have received the harmful build. BleepingComputer reports that the extension gathered system details, installed an SSH key controlled by the attacker, tried to switch on SSH access, and set up persistence. Researchers cited in that report spotted coding mistakes that likely stopped the backdoor on many Windows and macOS computers that did not have SSH turned on beforehand.
Cursor says the storage bucket held older coding-agent traces, mostly from 2025, from sessions run without Privacy Mode. Some traces contained API credentials that had not been scrubbed. By the company's figures, under 0.05% of users had traces exposed and under 0.01% of customers had API credentials exposed. With those credentials, the attacker reached some SaaS tools and took: archived non-production source repositories (some holding business contact details), limited support data and emails, usage metrics and contact details, summaries of some agent conversations, and certain credentials users had sent to support.
Cursor states that its production infrastructure, official app releases, customer code repositories, and the code and prompts of Privacy Mode users were not affected, and neither was Anthropic's genuine extension. The company revoked the key, rotated credentials, rebuilt the marketplace, moved all inference traffic away from Fireworks and closed its access to shared datasets. The Hacker News adds that Cursor reported no new malicious activity since November 1, 2026.
The links in this chain run through developer tooling
What stands out is that the way in was not Cursor's own app but a key exposed at a supplier, plus credentials sitting in logs kept as training data. BleepingComputer ties the incident to a wave of attacks on AI developer platforms that security firm Wiz calls "Black October". Per the report, those attacks aim mainly at credentials that open up AI models, and at Railway stolen secrets were later used for about $740,000 in fraudulent AI charges.
The Hacker News also writes that Uber disclosed a theft of driver license data after a service provider's credentials were compromised. Uber did not name the provider; the article says TechCrunch identified it as Fireworks AI, which is a media report and not something Uber confirmed.
What it means for businesses in Türkiye
The sources do not break the incident down by country, and this section is our reading, not a fact from them. For software teams, freelancers and agencies in Türkiye that use Cursor or another AI code editor, the lesson is general: an editor extension is part of the code you run, and an update from a marketplace is a supply-chain entry point.
Users who installed the affected builds, or updated to them, on October 29 are pointed to Cursor's own remediation guide. If you used those versions, checking the SSH keys and authorized-keys list on your machine and renewing the API keys you gave your editor or agents looks sensible; that is our suggestion, not the sources'. Reviewing the Privacy Mode setting also belongs on the list if you do not want your code and prompts used as training data.
The UNALSOFT view
The editors, extensions and API keys used while building a website or app are part of a project's security surface too. We read this story as a reminder to review the development environment and secret handling in web design projects. How much it matters to you depends on which versions you ran and where your keys live, so each team needs to check its own setup.
Have your development setup and secret keys been reviewed?
A short conversation is enough to go through your web project's structure and development process together.