Cursor security breach leads to malicious Claude Code extension
Hackers used stolen credentials to distribute a malware-infected extension on the Cursor marketplace, potentially affecting 5,000 developers.
The short version
- Attackers stole credentials from Fireworks AI to access Cursor's internal systems and customer data.
- Up to 5,000 developers may have downloaded a malicious extension containing persistent SSH backdoors.
- The breach exposed historical AI coding-agent traces, including unredacted API credentials for some users.
- Cursor has since revoked keys, rotated credentials, and rebuilt its extension marketplace.
Why it matters
This breach demonstrates how vulnerabilities in third-party AI infrastructure can cascade into developer tools. It serves as a warning for developers to monitor their extension installations and manage API keys carefully.
The breach and its origin
Cursor, an AI-powered code editor built on Microsoft's open-source Visual Studio Code, recently confirmed a significant security breach. The incident began when hackers obtained an access key from Fireworks AI, a third-party provider that Cursor uses for AI model training and inference. This initial theft allowed the attackers to move laterally into Cursor's internal systems.
Once inside, the attackers accessed a read-only storage bucket used to share information with Fireworks AI. This specific bucket contained historical AI coding-agent traces from 2025, which were collected from users who had not enabled Privacy Mode. Because these traces were not redacted, they contained sensitive information that the hackers were able to exploit.
The stolen data included unredacted API credentials, which the hackers used to compromise additional services. According to Cursor, the breach also allowed for the theft of archived, non-production source code repositories, customer support emails, and product usage metrics. The company confirmed that fewer than 0.01% of customers had their API credentials compromised in the incident.
Distribution of malicious code
The most critical impact of the breach occurred on October 29, when the attackers used a stolen credential to gain access to Cursor's extension marketplace. They published unauthorized copies of Anthropic's Claude Code extension directly onto the cache used to serve the marketplace. These malicious versions, specifically versions 2.1.312 and 2.1.313, remained available for approximately 11 hours before being removed by the company.
The hackers modified the legitimate extension to execute malicious JavaScript on the users' machines. This script was designed to install persistent SSH backdoors, which are methods for maintaining remote access to a computer. The malware also attempted to collect system information and configure persistence mechanisms to ensure the attackers could return to the infected devices at any time.
However, the effectiveness of the backdoor was limited by several coding errors. Security researchers noted that these errors likely prevented the backdoor from functioning on many Windows and macOS systems where SSH was not already enabled. Despite these technical hurdles, Cursor estimates that up to 5,000 users who requested the affected versions during the 11-hour window may have received the malicious builds.
Scope of the data exposure
The breach exposed a variety of data points, though Cursor noted that its production infrastructure and official application releases remained secure. The stolen information included summaries of some AI coding-agent conversations and credentials that customers had previously shared with the Cursor support team. The company emphasized that code and prompts belonging to users who had enabled Privacy Mode were not affected by the incident.
The breach is part of a broader trend of attacks targeting AI developer platforms, a period some cybersecurity firms have called "Black October." These attacks typically aim to steal authentication credentials to gain access to AI models for unauthorized inference or model distillation. Other companies, including Railway and OpenCode, reported similar incidents where stolen secrets led to fraudulent charges or exposed API keys.
Fireworks AI, the original source of the leaked key, also disclosed multiple security breaches during October. These included incidents where attackers accessed development machines through a compromised third-party cloud provider. One of these breaches also affected Notion, resulting in the exposure of AI chat transcripts from a small number of workspaces.
Remediation and next steps
Cursor responded to the breach by revoking the Fireworks access key and rotating all exposed credentials. To ensure the integrity of its platform, the company rebuilt its extension marketplace and redeployed its production infrastructure as a precautionary measure. They have also moved all inference traffic away from Fireworks AI to prevent further unauthorized access from that provider.
The company is currently investigating the full extent of the unauthorized access and has provided remediation guidance for those affected. Users who installed or updated to the malicious Claude Code extension during the exposure window are urged to follow these specific instructions to secure their systems. Cursor's monitoring systems were the primary mechanism that detected the unauthorized activity, allowing them to act quickly to remove the malicious content.
Moving forward, the incident highlights the risks associated with sharing data with third-party AI providers for model training. Developers are encouraged to be mindful of the data they share with AI agents and to keep Privacy Mode enabled whenever possible. The company continues to monitor for any further signs of unauthorized activity resulting from the stolen credentials.
What to watch
- How other AI infrastructure providers are hardening their access keys against lateral movement.
- Whether more developers will move toward 'Privacy Mode' by default to avoid data exposure in training traces.
Try it
Check your installed extensions in your code editor to ensure you are running the official version of Claude Code and not a malicious copy.
Sources and links
- Officialnew incident report cursor.com
- ReportingBleepingComputer: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by BleepingComputer · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
