← Back to blog Cybersecurity

Cursor breach leads to a trojanized Claude Code extension: how to check the IDE extensions and AI tools of your Mexico developers today

3 min read

Between November 14 and 15, 2026, Cursor, the AI code editor a lot of developers live in, confirmed a security incident: a stolen Fireworks AI key was used to publish a malicious Claude Code extension. Both BleepingComputer and The Hacker News covered it.

If your company has developers in Mexico, in-house, through an agency or a nearshoring partner, their code editor is a door almost nobody watches. And production keys usually sit right behind it.

What happened with Cursor and the Claude Code extension

An extension is an add-on installed inside the editor (VS Code, Cursor) to add features: autocomplete, an AI assistant, code formatting. It installs in one click and runs with the same permissions as the developer.

This is a supply chain attack. Nobody targeted one company in particular; they targeted a tool many companies trust. An API key (the password one program uses to talk to another service) ended up in the wrong hands and was used to slip in a trojanized version, meaning malicious code hidden inside something that looks legitimate.

In our experience, when we ask who knows which extensions the dev team has installed, the honest answer is nobody. Not even the team itself.

Why the code editor is such an attractive target

A developer laptop holds a bit of everything Source code, API keys, cloud tokens and sometimes the production database password.

The IDE (integrated development environment, the program where code gets written) lives on the developer laptop. That laptop usually holds files full of keys, open GitHub sessions, cloud access and direct connections to servers.

A malicious extension does not need to break anything. It just reads what is already there. That is a different angle from OAuth apps or shadow AI: here the risk sits inside the work tool itself.

For foreign companies this matters twice. Your Mexico dev team, or the nearshoring vendor you hired, may touch your repositories, your cloud and your ERP from machines your head office has never seen.

Wired made the broader point this week in its piece on the new AI-fueled ransomware boom: attackers look for shortcuts, and a forgotten key on a laptop is the perfect one.

What to do today, step by step

Inventory comes first You cannot clean up what you do not know is installed. One list per developer is enough to start.
ActionWhat it meansOwner
Inventory extensionsA list of VS Code and Cursor extensions per developerTech lead in Mexico
Pin versionsTurn off auto-updates and approve each new versionTech lead in Mexico
Rotate keys and tokensReplace API keys, GitHub and cloud tokens stored on devicesIT and dev team
Separate productionProduction credentials do not live on laptopsIT, head office
Review AI agentsWhat they can read, run and connect toTech lead in Mexico

Start with the inventory. If anyone has the affected extension, treat that laptop as compromised: take it off the network, rotate everything it touched and review recent access.

Keep production credentials off laptops

Production is not deployed from the couch Deployments run from an automated process, with keys in a secrets manager instead of loose files.

The rule we apply: developers work with test data and development keys. Production is deployed from a pipeline (an automated deployment process) with secrets stored in a manager, not in a file on someone’s desktop.

If your Mexico developer logs into the production database from a personal laptop today, it is not a crime. It is just the first thing we would change.

The AI agents in the editor have permissions too

Very helpful, very powerful An agent that runs commands without asking holds the same keys as your developer.

Claude Code, Cursor and similar tools no longer just suggest code. They run commands, read files and connect to other services. That saves hours, and it is also a huge permission.

Check which commands they can run without asking, which folders they can see and which external connections are configured.

Developer reviewing code at a workstation
Let us review your developer workstations with you Foto: Tirza van Dijk tirzavandijk · CC0 · Wikimedia Commons

What about your Mexico operation?

At ProcessBi we build custom software with these practices from day one: approved extensions, secrets off laptops and production kept separate.

And within our IT support plans we audit workstations and access in Mexico, including your developers, in-house or external, in English or Spanish.

Request a review of your developer workstations — we reply the same business day.

Your path

Running IT in Mexico from abroad

78 of 89
  1. Firewall and MSP ✓ Read You are here 3 min
  2. HR data breach ✓ Read You are here 3 min
  3. IDE extensions ✓ Read You are here 3 min
  4. Network changes ✓ Read You are here 3 min
  5. WhatsApp with AI ✓ Read You are here 3 min
  6. AI phone agents ✓ Read You are here 3 min
  7. Which AI to pay ✓ Read You are here 4 min
  8. Field services →
Next A misconfigured router grounded US flights: how to make network changes at your Mexico operation without taking it down