🤖 AI

A trusted Claude coding extension was altered on Cursor. Up to 5,000 users may have received it.

2 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
Anthropic(an-THROP-ik)

The company that makes Claude.

extension

A small program that adds features to another program.

software supply-chain incident

A security problem that travels through a trusted software delivery path.

Anthropic, the AI company behind Claude, had a modified version of its Claude Code extension distributed through Cursor, an AI coding tool. The altered package was available for nearly 12 hours on October 29. Cursor said the affected versions were 2.1.312 and 2.1.313. Ars Technica reported that up to 5,000 users may have been exposed. That is a possible reach, not a confirmed count of infected computers.

What happened

An extension is a small program that adds features to a coding app. A maliciously changed version appeared in Cursor’s VS Code Extension Marketplace. It came through an account and channel that looked official. Anthropic did not create the changed versions, according to Cursor’s notice. Ars reported that the attackers’ route into the verified Cursor account remained unknown.

Why the delivery route mattered

Cursor said the extension could collect basic information about a computer and its user. It also tried to leave a way for someone outside the computer to reconnect later. Those actions did not work in every case. But a coding extension can run close to a developer’s files and tools. That makes a trusted update path especially important. Many users do not inspect each update manually. Cursor says automatic updates are enabled by default. A single bad release can therefore reach many machines before anyone notices.

What has been confirmed

Cursor removed the altered versions and blocked further distribution. It briefly shut down its marketplace. The company rotated keys, checked access controls, and reviewed other extensions. It also worked with the hosting provider to take the attacker’s server offline. Cursor said it found no evidence that other extensions or the official distribution were altered. The marketplace later reopened.

Why it matters

This was not only a problem with one download. It was a software supply-chain incident. That means the danger traveled through a tool people trusted. For AI coding tools, the risk is wider than a broken feature. Developers may keep source code, account settings, and other work data nearby. A compromised add-on could create privacy and security work for both individuals and companies. The event also shows why a verified badge or official publisher name cannot be the only safety check.

What remains unclear

Several important facts are still missing. No public explanation has identified how the attackers gained access to Anthropic’s verified Cursor account. The report also did not establish how many people installed the package, how many devices executed its added code, or whether any information was taken. Ars said Anthropic had not publicly warned customers when it published its report. Cursor’s notice said the identified package was limited, but that does not answer every question about the original account access.

What to watch next

Readers should watch for a joint explanation of the access path, a confirmed user count, and clearer notice to affected customers. The broader industry question is how marketplaces verify publishers, sign updates, and respond when an official channel is abused. As AI coding assistants become more common, keeping the model safe will be only one part of the job. The software that delivers the assistant must be protected too.

Sources: Ars Technica and Cursor’s security notice.

🤖 AI

A bad version of Anthropic’s coding add-on reached Cursor

📰 Full story: A trusted Claude coding extension was altered on Cursor. Up to 5,000 users may have received it.

A coding tool shared a changed add-on for almost twelve hours.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
add-on

A small program that gives another program more features.

marketplace

A place where users get add-ons.

software supply-chain problem

A bad change that travels through a trusted delivery path.

💡 The gist

  • Anthropic, the company behind Claude, made the original add-on.
  • Cursor, an AI coding tool, shared a changed version.
  • Up to 5,000 users may have received it.

An add-on is a small program that gives another program more features. This add-on worked with coding tools. It appeared in Cursor’s marketplace. The marketplace is a place where users get add-ons.

The changed version was not made by Anthropic. It looked like an official release. Cursor said two versions were affected. The versions were 2.1.312 and 2.1.313. The bad version was available for nearly twelve hours.

The add-on tried to collect information about a computer. It also tried to leave a way to return later. It did not work the same way on every computer. Still, coding tools can reach important work files. Some users also receive updates automatically. That made the event more serious.

One reason is trust. People often trust a familiar publisher. They may not read every update. A bad update can move quickly before someone spots it. This kind of problem is called a software supply-chain problem. The danger comes through the delivery path. It is not only a problem inside the add-on.

Cursor removed the changed versions. It blocked new downloads. It closed the marketplace for a short time. It checked its access controls and other add-ons. It also took the related server offline. Then it reopened the marketplace.

We do not know how the attackers entered the verified account. We also do not know how many users installed the add-on. The 5,000 figure is a maximum possible reach. It is not a confirmed injury count. We do not know whether information was taken.

This story teaches one important lesson. An official name does not guarantee a safe update. The delivery system needs protection too. Companies should explain what happened next. They should also tell affected users clearly.

🤖 AI

A bad add-on reached a coding tool

📰 Full story: A trusted Claude coding extension was altered on Cursor. Up to 5,000 users may have received it.

Anthropic makes Claude, and its add-on got a bad version.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
Anthropic

The company that makes Claude.

Cursor

A tool that helps people write code.

add-on

A small program that adds extra abilities.

Anthropic is the company that makes Claude. Cursor is a tool that helps people write code. An add-on gives a computer program extra abilities.

A bad version appeared in Cursor. It stayed there for almost twelve hours. Up to 5,000 users may have received it.

Cursor removed the bad version. It closed the add-on store for a short time. Then it checked the store and opened it again.

We do not know how the bad version got there. We do not know how many computers used it. We do not know whether it took any information.

A familiar name does not always mean safety. The place that delivers software needs protection too.

Sources