CISO Voice

Breaches & Vulnerabilities / News

Cursor breach highlights the risks of third-party AI integrations

Attackers exploited credentials exposed through AI infrastructure provider Fireworks to access Cursor systems and potentially distribute malicious coding extensions to 5,000 users.

Ingrid Fadelli
Ingrid Fadelli Contributing Reporter · November 16, 2026 · 2 min read
Code open in an editor on a computer screen

AI coding platform Cursor has shared new details about a security incident in which attackers used credentials exposed through third-party AI infrastructure provider Fireworks to access internal systems and distribute malicious developer extensions.

In an update published on Nov. 13, Cursor said attackers initially used an exposed Fireworks access key to copy data from a shared storage bucket containing coding-agent traces and some unredacted credentials.

The attackers subsequently used these credentials to access internal SaaS services, obtaining customer support information, email conversations, usage metrics and archived non-production source code.

On Oct. 29, the company revealed that attackers had added malicious versions of Anthropic’s Claude Code extension to its marketplace cache. The extensions were available for approximately 11 hours, and Cursor said in its latest update that up to 5,000 users could have received them.

Microsoft researchers found that the extensions contained code designed to establish persistent remote access through unauthorized SSH keys and reverse SSH tunnels, although the backdoor may not have been established successfully on all affected devices.

The incident at Cursor follows a separate disclosure by collaboration platform Notion, which reported the potential exposure of users’ AI chat histories through services linked to Fireworks.

Impact and remediation measures

Cursor said fewer than 0.05% of users had historical coding-agent traces exposed, while fewer than 0.01% of customers had API credentials exposed. Its production systems, official application and software updates were unaffected.

Since the breach, the company has rotated credentials, rebuilt its marketplace store, strengthened security controls and discontinued its use of Fireworks.

“This incident involved data we stored and shared, credentials we issued, and systems we operate,” Cursor said. “We are addressing the impacts and are enhancing our security measures based on what we have learned.”

Cursor urged users who installed affected Claude Code extension versions 2.1.312 or 2.1.313 to check for signs of compromise, warning that uninstalling the extensions alone may not remove persistent malicious components.

The company said it has notified affected customers and believes the incident has now been contained.

Ingrid Fadelli
Ingrid Fadelli Contributing Reporter

Ingrid Fadelli is a freelance journalist with a BSc in Psychology and an MA in International Journalism, both from City University London. For the past 10 years, she has been writing articles focusing on research and emerging technologies in various fields, including AI, robotics, electronics engineering, cybersecurity, neuroscience, biology, physics and environmental science. She published articles in Phys.org, TechXplore, MedicalXpress, Scientia, E&T Magazine, and on various other media, translating complex scientific developments into engaging stories for broad audiences.

Keep reading for free,
one signup unlocks everything

We'll send you the CISO Voice newsletter. Unsubscribe anytime.

More from CISO Voice

Sign up to CISO Voice, the free newsletter trusted by senior security leaders.

Secret Link