Home › Intelligence › Brief
BREACH BRIEF 🟠 High Breach

Cursor Breach Exposes Developer Credentials via Stolen Fireworks AI Key, Leads to Malicious Claude Code Extension

Cursor disclosed that attackers stole an access key from third‑party provider Fireworks AI, harvested unredacted API credentials, and used them to publish malicious Claude Code extensions on its marketplace, potentially affecting up to 5,000 developers. The breach highlights the importance of third‑party access oversight and continuous credential monitoring for audit readiness.

Verisq™ Intelligence · 📅 November 15, 2026 · 📰 bleepingcomputer.com
Get the weekly threat digestBreaches, exploited vulnerabilities and advisories that matter to your vendors, every Friday. Free.
🟠
Severity
High
BR
Type
Breach
🎯
Confidence
High
🏢
Affected
2 sector(s)
✅
Actions
4 recommended
📰
Source
bleepingcomputer.com

Cursor Breach Exposes Developer Credentials via Stolen Fireworks AI Key, Leads to Malicious Claude Code Extension

What Happened — Cursor, an AI‑powered code editor built on Microsoft VS Code, disclosed a supply‑chain breach that began with the theft of an access key from its AI‑infrastructure partner, Fireworks AI. On October 29 2026, attackers used the stolen key to copy data from a read‑only storage bucket that Cursor shared with Fireworks. The bucket contained historical AI‑coding‑agent traces from 2025, many of which included unredacted API credentials because the users had not enabled Privacy Mode.

With those credentials, the threat actors accessed internal Cursor services, stole archived source‑code repositories, limited customer‑support emails, usage metrics, and other non‑production data. Crucially, they also obtained a privileged credential that allowed publishing extensions to Cursor’s VS Code marketplace. The attackers uploaded malicious copies of Anthropic’s Claude Code extension (versions 2.1.312 and 2.1.313).

The trojanized extensions were available for roughly 11 hours before Cursor removed them. When installed, the malicious JavaScript attempted to install persistent SSH backdoors on developers’ machines, collecting system information and planting attacker‑controlled SSH keys. Security researchers noted coding errors that likely prevented the backdoor from functioning on many Windows and macOS hosts where SSH was not already enabled, but the risk remained significant.

Cursor estimates that up to 5,000 developers who requested the affected extension during the exposure window may have received the malicious build. Less than 0.05 % of users had coding‑agent traces exposed, and fewer than 0.01 % of customers had API credentials compromised, but the absolute numbers still represent a notable breach of trust.

The company’s monitoring systems flagged the unauthorized activity, prompting immediate revocation of the Fireworks access key, rotation of all exposed credentials, and a temporary shutdown of several internal services. Cursor also rebuilt its extension marketplace, migrated inference traffic away from Fireworks, and redeployed production infrastructure as a precaution.

In parallel, the incident is part of a broader wave of attacks dubbed “Black October,” where threat actors target AI developer platforms to harvest authentication secrets for illicit AI inference and model‑distillation activities. Other victims cited include Railway, Northflank, Lightning AI, OpenCode, and Warp. Fireworks AI itself disclosed multiple breaches in October, affecting downstream customers such as Notion.

Cursor has issued remediation guidance for any developer who installed the malicious Claude Code extension and is cooperating with external investigators. The company emphasizes that its production code releases, customer code repositories, and any data belonging to users with Privacy Mode enabled were not impacted.

Why It Matters for Trust & Control Assurance — The incident underscores the need for rigorous third‑party access oversight and continuous credential‑monitoring to protect supply‑chain integrity.

Who Is Affected — SaaS AI‑developer platforms, their enterprise development teams, and any downstream services that integrate with Cursor’s marketplace.

Recommended Actions

  • Review and tighten third‑party access policies; enforce least‑privilege for any external keys.
  • Implement continuous monitoring of credential usage and automated rotation for all third‑party secrets.
  • Conduct a rapid audit of extension marketplace publishing rights and enforce code‑signing verification.
  • Advise developers to uninstall the affected Claude Code versions, rotate SSH keys, and verify system integrity.
  • Update privacy controls (e.g., enable Privacy Mode) to prevent accidental credential leakage in AI‑agent logs.

Technical Notes

  • Attack vector: Compromise of a third‑party provider’s access key (Fireworks AI) → unauthorized read‑only bucket access → credential harvest → malicious marketplace publishing.
  • Malware behavior: JavaScript payload attempts SSH key injection, persistence via system services; limited success on Windows/macOS due to missing SSH daemon.
  • Data exposed: Archived non‑production source code, support emails, usage metrics, AI‑coding‑agent conversation summaries, API keys.
  • Timeline: Breach discovered via internal monitoring → key revocation & credential rotation → marketplace takedown (≈ 11 h exposure).

Source: BleepingComputer

📰 Original Source
https://www.bleepingcomputer.com/news/security/cursor-confirms-security-breach-led-to-malicious-claude-code-extension/ ↗

This Verisq Intelligence Brief is an independent analysis. Read the original reporting at the link above.

Third-party risk

Does this breach reach you?

Verisq continuously monitors your vendors for breach and ransomware activity, so the question stops being whether it happened and becomes whether it reaches you.

See a live Trust Center →