Skip to content
Threat Feed
high threat exploited

Cursor: Supply chain compromise via Fireworks AI breach

Attackers leveraged a compromised access key from Fireworks AI to gain unauthorized access to Cursor internal systems and inject malicious copies of the Anthropic Claude Code extension into the Cursor VS Code marketplace cache.

Triage facts

Revision 1
Applicability
anthropic.claude-code extension (versions 2.1.312 and 2.1.313)
Evidence
Confirmed activeMedium confidence
Severity
High
Sources
1 public source
Freshness
Revision 1Nov 16, 17:10 UTC
Source quality
C2Likely

Overview

In October 2026, the AI inference platform Fireworks AI experienced two security incidents. The first involved the theft of internal credentials, and the second involved unauthorized access to development machines via a third-party cloud provider. During these incidents, a Cursor access key held by Fireworks was compromised.

Attack Chain

Using the compromised access key, attackers accessed Cursor's internal systems, including non-production source code repositories, customer support data, and credentials. Following this, the attackers used exfiltrated secrets to place unauthorized, malicious copies of the 'anthropic.claude-code' extension (specifically versions 2.1.312 and 2.1.313) into the cache used to serve the Cursor VS Code extension marketplace. These malicious versions were served to users for approximately 11 hours. Cursor discovered the activity on October 29 and blocked the attacker's access.

Impact

Cursor estimates that at most 5,000 users were affected by the malicious extension copies. Accessed data included non-production source code, customer support emails and conversations, product usage metrics, and user credentials submitted to the Cursor support team.

Recommendation

Cursor considers the incident contained as of November 1. Users who installed or updated to 'anthropic.claude-code' versions 2.1.312 or 2.1.313 should follow the remediation steps provided in the official Cursor security notice, as simple uninstallation may be insufficient.

Threat Hunt

Monitor authentication logs for API keys or service tokens associated with third-party partners/vendors that are accessed from anomalous IP addresses or used to perform unusual cross-bucket read operations.

T1199 high high confidence hunt now

Data: Cloud provider API access logs, Identity and Access Management (IAM) audit logs

Search for scripts or automated processes searching for common credential file paths (e.g., .env, config.json, ssh keys) within non-production or archived source code repositories.

T1552 high medium confidence hunt now

Data: Endpoint detection and response (EDR) process execution logs, Data Loss Prevention (DLP) logs for file scanning activity

Baseline and monitor integrity of artifact caching servers or update distribution points; alert on unauthorized modifications or the replacement of signed binaries/extensions with unsigned or differently-signed versions.

T1195 high high confidence convert to detection

Data: File integrity monitoring (FIM) logs, Content Delivery Network (CDN) / Cache server access logs