18 Nov 2026, Wed

Cursor Supply Chain Attack: Compromised Fireworks AI Credentials Lead to Malicious Claude Code Extensions

In a significant supply chain security breach, the AI-powered code editor Cursor recently fell victim to an sophisticated attack that allowed threat actors to distribute malicious versions of the Anthropic claude-code extension to its user base. The incident, which highlights the cascading risks inherent in modern software development ecosystems, originated from the compromise of credentials held by the AI inference platform Fireworks AI. This unauthorized access enabled attackers to infiltrate Cursor’s internal systems, exfiltrate sensitive data, and ultimately inject malicious code into the extension marketplace cache.

The breach was officially identified by Cursor on October 29, triggering an urgent security response. During an approximately 11-hour window, two modified versions of the anthropic.claude-code extension—specifically versions 2.1.312 and 2.1.313—were made available to users through the Cursor VS Code Extension Marketplace. It is important to emphasize that Anthropic, the developer of Claude, was not the target of this attack and suffered no breach of its own infrastructure. The entire malicious campaign occurred within the environment managed by Cursor, stemming from a security failure that had its roots elsewhere in the supply chain.

The Chain Reaction: From Fireworks AI to Cursor

The genesis of the incident lies in a series of security failures at Fireworks AI, which suffered two distinct and impactful breaches throughout October. According to disclosures from both companies, these incidents provided the initial foothold for the attackers. In the first of these events, threat actors successfully utilized internal Fireworks credentials to gain unauthorized access to stored secrets, alongside sensitive account and billing information.

The situation escalated significantly when a second, separate incident involved the compromise of a third-party cloud provider used by Fireworks. This intrusion allowed the attackers to move laterally into Fireworks’ development machines, resulting in the theft of customer data. During this chaotic period of compromised infrastructure, a specific access key that had been issued by Cursor to Fireworks—intended for legitimate operational integration—was also exposed.

The attackers wasted no time in weaponizing the stolen Cursor access key. By utilizing this key, the threat actors were able to bypass standard security layers and access a read-only storage bucket belonging to Cursor. Crucially, this bucket contained unredacted credentials that acted as a "skeleton key" for various SaaS (Software as a Service) tools that Cursor relies upon for its daily operations. With these credentials in hand, the attackers moved from the external storage bucket into the heart of Cursor’s internal systems.

Extent of the Internal Breach

Once inside, the impact of the intrusion was extensive. Cursor’s internal investigation revealed that the attackers had successfully accessed a range of sensitive information. This included select archived, non-production source code repositories, some of which contained business contact information. Furthermore, the attackers managed to harvest limited customer support data, including historical email conversations, product usage metrics, and specific contact details.

Perhaps most concerning was the discovery that the attackers had accessed summaries of certain coding agent conversations and, in some instances, credentials that users had proactively sent to the Cursor support team for troubleshooting purposes. This level of access underscores the inherent dangers of credential sprawl, where internal keys and sensitive tokens, if not properly secured or rotated, can provide an attacker with a comprehensive view of a company’s internal operations and its user data.

Cursor moved to block the attacker’s access immediately upon discovery of the unauthorized activity. However, the damage had already been done. Before they were evicted from the system, the attackers leveraged the secrets exfiltrated from these internal tools to orchestrate a supply chain attack. They successfully placed malicious, unauthorized copies of Anthropic’s Claude Code extension directly onto the cache used to serve the Cursor VS Code extension marketplace.

Impact on Users and Mitigation Efforts

The nature of the attack meant that the malicious software was never present in the Open VSX marketplace source, which serves as a primary repository for many VS Code extensions. Instead, the threat was isolated to the specific cache used by Cursor. Consequently, only those users who requested the cached versions of the extension during the 11-hour window of exposure were at risk. Cursor estimates that at most 5,000 users were affected by this specific tampering.

For those users who may have installed or updated to the tampered versions (2.1.312 or 2.1.313), the situation is serious. Cursor has issued a formal security notice urging these users to take immediate, specific remediation steps. The company warns that simply uninstalling the malicious extension is insufficient to ensure security, as the compromised credentials or tokens that might have been handled by the extension could still be active or stored elsewhere in the user’s environment. Customers are strongly encouraged to review the company’s official security notice for guidance on revoking keys and cleaning their development environments.

By November 1, Cursor reported that it had successfully contained the incident and had seen no further evidence of attacker activity within its environment. The company continues to monitor its systems closely, reflecting a broader trend in the tech industry where even robustly designed platforms must contend with the vulnerabilities of their upstream partners.

A Broader Pattern of Supply Chain Vulnerabilities

The incident involving Cursor is not an isolated event but rather part of a troubling series of breaches linked to the ongoing security struggles at Fireworks AI. The ripple effects of the Fireworks compromise have been felt across the industry, with companies like Notion and Uber also confirming that they were impacted by data breaches stemming from the same underlying security issues at the inference platform.

These events serve as a sobering reminder of the interconnectedness of modern digital infrastructure. When a service provider or a third-party partner experiences a security failure, the blast radius of that incident can extend rapidly to their clients, partners, and customers. For development platforms like Cursor, which rely on a dense web of APIs, cloud services, and third-party extensions, the challenge of maintaining a secure supply chain is increasingly complex.

The incident highlights a fundamental shift in the threat landscape: attackers are no longer just targeting primary, high-value entities directly. Instead, they are increasingly focused on the "soft underbelly" of the supply chain—the smaller integrations, the shared access keys, and the trusted developer tools that provide a gateway into larger, more fortified networks. The use of a stolen, read-only storage key to pivot into a company’s core SaaS tools is a textbook example of how a relatively minor initial compromise can be escalated into a significant supply chain attack.

As the industry grapples with these revelations, the focus for many firms is shifting toward "zero trust" architectures, more rigorous credential management, and the implementation of automated supply chain integrity checks. The fact that the malicious extensions were injected into a cache, rather than the primary source, suggests that traditional security audits might miss subtle, transient threats that exist only in the distribution layer.

For the developer community, the Cursor incident serves as a call to vigilance. While tools like VS Code extensions offer immense productivity benefits, they also introduce a vector that can be exploited in ways that are difficult for the end user to detect. As companies continue to tighten their security postures in the wake of these disclosures, the need for transparent communication—as demonstrated by Cursor’s prompt disclosure—remains the most effective way to limit the long-term impact on user trust and system integrity. As of the current reports, the situation is considered contained, but it remains a definitive case study in the risks of the modern software supply chain.

Leave a Reply

Your email address will not be published. Required fields are marked *