On 13 November 2026, Cursor published a detailed update on a security incident that first became public on 29 October 2026. The company said the vast majority of customers were not affected, and that it has contacted each affected customer with the details that apply to them. Even so, the story matters to almost any team that writes software with an AI coding assistant, because it started with a vendor access key and later reached a cached copy of a popular editor extension.
This is not a bug in Claude Code itself, and it is not the same as earlier reports about secrets sitting in local Claude settings files. It is a vendor-key and marketplace-cache incident. If your developers use Cursor, or if your company issues access keys to AI vendors, you should understand what was exposed, what was not, and what to do next. DayDecoder is a custom software company in Ahmedabad, and we use AI coding tools in client work. Incidents like this are why those tools need the same access rules you would give any other production system.
Key takeaways
- Cursor’s 13 November 2026 update says the incident began when an access key it had issued to Fireworks was exposed by that vendor and used to copy data from a shared storage bucket.
- A small share of coding-agent traces, mostly from 2025 and only from sessions not in Privacy Mode, sat in that bucket. Cursor said less than 0.01% of customers had API credentials exposed in those traces.
- On 29 October 2026, the attacker placed malicious copies of the Claude Code extension, versions 2.1.312 and 2.1.313, on the cache for Cursor’s VS Code extension marketplace. The copies were available for about 11 hours. At most 5,000 users who requested those specific cached versions may have received them.
- Anthropic did not publish those copies, and Cursor said this was not a vulnerability in Anthropic’s product. The usual marketplace source, Open VSX, never had the malware.
- Cursor said prompts and code from Privacy Mode users were not affected, production Cursor systems were not affected, and the last attacker activity it has seen was on 1 November 2026 and was blocked. If you may have installed those two versions from the Cursor marketplace during the window, follow Cursor’s official security notice. Uninstalling the extension alone may not be enough.
What happened, in order
Cursor’s investigation, published on 13 November 2026 after about two weeks of work, describes the incident in five stages. The dates below are from that official update.
It started at the vendor. Cursor had issued an access key to Fireworks and used a read-only storage bucket to share data with that vendor. The bucket held limited coding-agent traces, mostly from 2025, from sessions that were not in Privacy Mode, which Cursor said it was permitted to use for model training. A small number of those traces contained credentials that had not been redacted. Cursor said less than 0.01% of customers had API credentials exposed in this way. Using some of those credentials, the attacker then reached selected SaaS tools used by Cursor. From those tools, Cursor said the attacker accessed selected archived, non-production Cursor source-code repositories, some of which contained business contact information; limited customer-support data and email conversations; product-usage metrics and contact details; summaries of some coding-agent conversations; and some credentials that users had sent to the Cursor support team. That last point is a reminder for every team: do not paste live API keys, passwords or private certificates into a support ticket if a redacted extract or a rotated test key would do.
Cursor detected unauthorised activity and began cutting off access. After it became aware of the activity, Cursor deactivated the vendor’s key, opened an incident-response effort, and began revoking and rotating credentials. As further access was found, it was cut off. Out of caution, Cursor also temporarily disabled internal and product services beyond what the evidence required, which briefly disrupted some Cursor features. Those services were restored once Cursor judged it safe to do so.
A malicious extension was placed on the marketplace cache. On 29 October, using a stolen credential from the exposed internal data, the attacker put malicious, unauthorised copies of Anthropic’s Claude Code extension directly on the cache used to serve the Cursor VS Code extension marketplace. Cursor is clear on two points that are easy to get wrong. Anthropic did not publish this copy, and this was not a vulnerability in Anthropic’s product. The marketplace source, Open VSX, never had the malware. The problem was the cache in front of Cursor’s marketplace, not the official upstream listing. Cursor said that at most 5,000 users who requested specific cached versions on 29 October may have received the malicious build, and that it was available for about 11 hours before Cursor removed it. Cursor then took the extension marketplace offline and worked with the cloud provider hosting the attacker’s server to take that server down.
Customers were notified. On 30 October, Cursor published a security notice with indicators of compromise and remediation steps, and began notifying affected users and team admins in the app and by email. If credentials were exposed, Cursor said it emailed team admins a list of the specific keys to rotate, and asked the third-party services that issued those credentials to revoke them. If customer-support or business data may have been involved, Cursor said it notified those customers directly. If someone installed or updated to the tampered versions Anthropic.claude-code 2.1.312 or 2.1.313 from the Cursor VS Code extension marketplace during the exposure window, Cursor asks them to follow the steps in that official security notice. Uninstalling the extension alone may not be enough.
Cursor believes the incident is contained. The last attacker activity Cursor has seen was on 1 November 2026, and it was blocked. The company said it continues to monitor its systems and to strengthen controls, and that it will contact customers directly if it finds additional information that affects them.
What was affected, and what was not
This section is easy to compress into a rumour, so it is worth keeping Cursor’s distinctions intact. Two different percentages appear in the official update, and they describe two different things.
| Question | What Cursor reported on 13 November 2026 |
|---|---|
| Prompts and code in Privacy Mode | Not affected |
| Coding-agent traces, for users not in Privacy Mode | Exposed for fewer than 0.05% of users. Cursor said it identified and notified each of those users. If you did not receive a notification, Cursor said your data was not in these traces |
| API credentials inside those traces | Exposed for less than 0.01% of customers |
| Live traffic you send to Cursor now | Not affected. The exposed traces were limited to storage used for historical offline analysis |
| Production systems that run Cursor | Not affected |
| The Cursor app, official updates, and release signing keys | Not affected. Cursor said it ships the app from a separate code-hosting system that the attacker never accessed, and that no release signing keys were exposed |
| Other marketplace extensions and plugins from the Cursor Marketplace | Not affected. Cursor said it validated that all other extensions were authentic, and reviewed and rebuilt the marketplace store |
| Customer code repositories | Not affected |
| Active inference requests at the time | Not exposed. Cursor later moved all inference away from the third-party vendor as a precaution |
| Malicious Claude Code copies on the Cursor marketplace cache | Versions 2.1.312 and 2.1.313, for about 11 hours on 29 October, potentially reaching at most 5,000 users who requested those specific cached versions |
Cursor also said its other products, including Cloud Agents for customer teams, were not affected. The archived, non-production repositories and the support and metrics data described earlier are separate from production source and from customer git repositories. If you use Privacy Mode, Cursor said your prompts and code were not affected. Privacy Mode is not a full security programme, but in this incident it was the difference between traces that could be shared with a vendor and traces that were not in that bucket.
What Cursor says it has done
The 13 November update lists the steps Cursor has already taken. It moved all inference traffic off Fireworks, deactivated the key it had issued to that vendor, and cut off the vendor’s access to all data sets Cursor had shared with it. It rotated every exposed Cursor credential it found, highest risk first, and, as a precaution, rotated all credentials in use across its internal systems, even where it had no evidence of exposure. It also said it validated every other marketplace extension, reviewed and rebuilt that store, expanded network-access restrictions, added secret scanning, redeployed production infrastructure as a precaution, reviewed internal services before restoring them, engaged external forensics experts, and contacted everyone affected.
Those are Cursor’s statements about its own environment. They do not, by themselves, clean a developer laptop that installed a tampered extension during the exposure window. That part is still on the affected user and their employer, following Cursor’s official notice.
What your team should do now
You do not need a full forensic exercise for every developer. You do need a short, written check, especially if people in your company use Cursor with Claude Code.
- See whether Cursor has already contacted you. If credentials were exposed, Cursor said it emailed team admins a list of specific keys to rotate. If support or business data may have been involved, it said it notified those customers directly. If you received that message, rotate the listed keys, confirm they have been revoked, and record the time you did it.
- Check Claude Code versions on machines that update from the Cursor VS Code extension marketplace. Cursor named versions 2.1.312 and 2.1.313, during the exposure window on 29 October 2026. If a machine may have installed or updated to either version from that marketplace in that window, follow Cursor’s official security notice. Uninstalling the extension alone may not be enough.
- Rotate secrets that may have been pasted into agent sessions or support tickets, even if you were not on Cursor’s email list, if your own logs show that a key was shared that way. Then put secrets in a manager, not in chat, tickets or unredacted traces.
- Turn on Privacy Mode for company workspaces if you have not already, and write down which projects are allowed to leave it off. In this incident, traces from Privacy Mode sessions were not in the vendor bucket.
- Review the access keys you issue to AI vendors. A key that can read a training or trace bucket is still a production secret. Give it the narrowest possible access, watch for unusual reads, and be ready to revoke it in minutes. Do not assume a “read-only” bucket is harmless if the files inside can contain credentials.
- Treat editor marketplaces and their caches as part of your supply chain, and use only Cursor’s official notice for cleanup. Unverified scripts from unofficial posts can make an incident worse. “We install from the usual marketplace” is not the same as “we install a verified build.”
If you did not receive a notification, Cursor said your data was not in the exposed traces. That statement covers the trace bucket. It does not automatically answer whether a machine installed a tampered extension version. Those are two different checks.
Lessons for teams that use AI coding tools
The useful lesson is not “stop using coding assistants.” It is that these assistants now sit on the same kind of path as CI systems and cloud accounts: they hold traces, they receive secrets by mistake, they install extensions, and they often have keys into other services.
Three habits would have reduced the harm. First, redact secrets before any trace, log or training example leaves the machine or the product. Cursor said a small number of traces still contained credentials that had not been redacted. Second, keep a hard line between production source and any archive or vendor share, and keep support tickets free of live credentials. Third, assume that a key issued to a vendor will eventually be mishandled, and design it so the damage is limited and visible.
Keep this incident separate from earlier reports about secrets in local Claude settings files. That problem lives in a developer’s own repository and editor config. This November update is about a vendor key, a shared trace bucket, internal support tools, and a marketplace cache. The practical advice overlaps, but the cause is different.
How we can help at DayDecoder
DayDecoder is a custom software company in Ahmedabad. We build custom web and mobile products, and we help teams adopt AI coding tools without treating those tools as a back door into production. If you want a practical review of how your developers use Cursor, where secrets can leak, and how to tighten vendor access, use the contact form. We reply within one business day.
FAQ
When did this happen?
Cursor first said on 29 October 2026 that it was investigating, and that a malicious copy of the Claude Code extension had appeared in its VS Code extension marketplace. It published a fuller investigation update on 13 November 2026. It said the last attacker activity it has seen was on 1 November 2026 and was blocked.
Was this a flaw in Claude Code or in Anthropic?
Cursor said Anthropic did not publish the malicious copy, and that this was not a vulnerability in Anthropic’s product. The official upstream marketplace source, Open VSX, never had the malware. The malicious copies were placed on the cache used to serve Cursor’s own marketplace.
I use Privacy Mode. Is my code safe?
Cursor said prompts and code from Privacy Mode users were not affected. For people not using Privacy Mode, it said coding-agent traces were exposed for fewer than 0.05% of users, and that those users were notified. If you did not receive a notification, Cursor said your data was not in those traces.
Which extension versions should we look for?
Cursor named tampered versions Anthropic.claude-code 2.1.312 and 2.1.313, installed or updated from the Cursor VS Code extension marketplace during the exposure window on 29 October 2026. If that may apply to a machine in your team, follow Cursor’s official security notice. Uninstalling the extension alone may not be enough.
Were customer git repositories stolen?
Cursor said customer code repositories were not affected, and that the production Cursor app and its official updates were not affected. It did report access to some archived non-production Cursor repositories and to some support and business data for notified customers.
Written by
Prakash Tank
Founder & Full-Stack Architect, DayDecoder
Prakash started as a certified Laravel developer and grew into a full-stack architect who owns a project end to end: requirements, architecture, development, deployment, and support. He designs event-driven systems with...
More about Prakash