Cursor has officially confirmed that it suffered a security breach, because clearly, letting hackers waltz in and distribute malicious software is the new normal. In this latest tech debacle, hackers snagged credentials from the AI infrastructure provider Fireworks AI and used them to promote a malicious version of the Claude Code extension through Cursor's marketplace.
The breach didn't just stop at distributing bad software; it also gave attackers a glorious backstage pass to steal customer data and access internal systems. Among the treasures pilfered were historical coding traces that included unredacted credentials, nicely wrapped for further exploitation.
The incident report, released just yesterday, outlines how the attackers struck on October 29, managing to distribute the malicious extension without breaking a sweat. Apparently, Cursor's marketplace is as secure as a screen door on a submarine.
These hackers had the audacity to expose around 5,000 developers to malware that installed persistent SSH backdoors. Someone really ought to remind these companies that securing user data is kind of a thing.
The breach kicked off when hackers discovered an access key carelessly stored within Fireworks AI, the third-party provider that assists with AI model training. They used this key to filch data from a read-only storage bucket shared with Fireworks, which contained unprotected coding traces dating back to 2025. And no, not all users were given the option to protect their data with Privacy Mode. Lucky them.
Cursor insists that only a minuscule fraction—0.05%—of users had their coding traces exposed, and even fewer, 0.01%, had credentials compromised. To add to the irony, the stolen data included benign details like archived source code repositories and customer support emails that no one outside of Cursor probably cares about.
Cursor's monitoring systems did detect some of the unauthorized activity, allowing them to revoke the Fireworks access key and rotate the exposed credentials. But let's not forget that the attackers had already made merry with a credential enabling them to distribute extensions on the marketplace.
For a brief period, the malicious versions of the Claude Code extension were available for around 11 hours before anyone thought to pull them from the shelves. The malware was crafted to execute sinister JavaScript that attempted to install persistent SSH backdoors on developers' machines. Hope everyone had their shields up.
Sadly, security researchers found coding errors in the malware, which probably saved many Windows and macOS users from this particular disaster. Because preventing attacks through design flaws is what we call a fortunate break in this reckless game of digital tag.
Cursor claims their production infrastructure and customer repositories were unharmed, especially those users who had the foresight to enable Privacy Mode—because nothing screams "I trust this platform" like having to actively safeguard your own data.
In an ironic twist, this incident comes on the heels of a cybersecurity wave dubbed "Black October," where various AI developer platforms faced similar attacks. Apparently, it's an excellent month for hackers looking to cash in on stolen authentication credentials.
So here we are, watching yet another tech company trip over its own incompetence. What’s next? Another round of "Oops, we did it again" from Silicon Valley? Enjoy the ride, folks.

