🔥 Trending on HN

A Tampered Claude Code Extension Entered Cursor’s Store

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

Words
extension(ik-STEN-shən)

An add-on that gives a larger program extra abilities.

SSH backdoor(S-S-H backdoor)

A hidden way to enter a computer remotely.

supply-chain attack(supply-chain attack)

An attack on the route used to deliver trusted software.

What happened

On October 29, 2026, Cursor, a code-editing application, distributed a tampered Claude Code extension through its extension marketplace. Claude Code is made by Anthropic, the AI company behind Claude. The affected versions were 2.1.312 and 2.1.313. Cursor said they were available from about 2:27 a.m. to 1:35 p.m. Pacific Time. It also said Anthropic did not develop those malicious versions.

The incident was first documented by Phorion in a security advisory. Cursor published an official notice. Microsoft later published its own technical analysis.

The supplied Hacker News data show strong community attention. One post had 829 points and 504 comments. Another had 654 points and 17 comments. Those numbers measure discussion, not the truth of the original report. The posts are here and here.

Why an extension mattered

An extension is an add-on for a larger program. It runs inside the editor and can use the editor’s permissions. Phorion said such an add-on may reach local files, source code, and developer tools. That makes the distribution channel part of the security boundary.

A package can look official while being changed somewhere between its creator and its users. This is a software supply-chain attack. The target is not only the application itself. The target is the path used to deliver trusted software.

What the package tried to do

The modified extension tried to collect basic device and user information. It also tried to add an attacker-controlled SSH key, prepare remote access, create a reverse connection, and leave settings that could run again later. In simple terms, it tried to create an SSH backdoor, or a hidden way into a computer.

Microsoft said the package also tried to send environment-variable names. It did not find evidence that the values of those secrets were stolen. Success depended on each computer’s permissions and settings. Cursor said errors and missing administrator rights blocked remote access on most macOS and Windows computers. That did not necessarily prevent the package from leaving keys or other traces.

What has been confirmed

Cursor removed the affected versions and temporarily shut down its marketplace. It audited the other extensions, rotated its keys, rebuilt the marketplace, and brought it back online. Cursor also worked with the hosting provider to take the attacker’s server offline.

Microsoft said the malicious packages were not published through Microsoft’s Visual Studio Marketplace or Anthropic’s official npm distribution. Anthropic also told Microsoft that its systems were not compromised. Current evidence points to the Cursor marketplace as the affected distribution channel.

What remains unknown

Public information does not establish how many people installed the versions. It also does not show how many machines executed the malicious code, how many gained successful remote access, or whether confidential data was taken. The attacker’s server being offline limits continued contact. It does not automatically remove files or settings already left on a computer.

Security teams therefore need to identify machines that installed or updated to the affected versions. They may need to preserve evidence, remove confirmed artifacts, and rotate credentials when exposure is indicated. Uninstalling the extension alone may not be enough.

What to watch next

The next important facts will be the number of affected machines, evidence of real-world access, and any further statements from Cursor or Anthropic. The wider lesson concerns trust in AI development tools. Model safety matters, but so do extensions, marketplaces, update systems, and the checks behind them. This incident shows that a trusted name is not enough. Users also need confidence in the route that delivers the software.

💬 The fix was welcomed, but the root cause and responsibility remain unclear

HN commenters generally viewed Cursor’s cleanup guidance positively, while questioning the intrusion path, proof of prevention, and the division of responsibility between Anthropic and Cursor.

  • One commenter described Cursor’s security notice as a solid response because it included cleanup instructions and scripts, but said it did not explain how the infection entered.
  • Another agreed that there was no root-cause analysis and asked how the response could demonstrate that the problem would not recur, framing it as a possible security-culture failure rather than an isolated incident.
  • Some interpreted Cursor’s statement that Anthropic did not develop the malicious extension as meaning that no Anthropic employee intentionally inserted an SSH backdoor.
  • Others criticized that wording as vague and evasive, arguing that it could obscure the fact that the extension was published on the Cursor marketplace under Anthropic’s name.
  • A counterargument noted that taking down and rebuilding the marketplace registry could indicate that Cursor’s infrastructure was the intrusion source, so attribution should not be rushed. The commenter later edited the post to say subsequent information supported that view.
  • Another commenter pointed out the irony of the notice linking to news that Cursor had earned AIUC-1 certification for agent security and reliability.

initial digest at 17 comments (revision 1). We fetched 17 comments and sampled 17 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

A Bad Claude Code Copy Appeared in Cursor’s Store

📰 Full story: A Tampered Claude Code Extension Entered Cursor’s Store

A changed coding tool appeared in Cursor’s extension store and tried to enter computers.

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

Words
extension(ik-STEN-shən)

An extra tool added to a computer program.

backdoor(BACK-door)

A hidden way into a computer.

marketplace(MARK-it-place)

A store where software add-ons are offered.

💡 The gist

  • Anthropic makes Claude and Claude Code.
  • A changed Claude Code extension appeared in Cursor’s store.
  • It tried to leave a hidden way into computers.

Anthropic is the AI company behind Claude. Claude Code helps people write software. Cursor is a code editor. Its marketplace offers extra tools called extensions.

An extension is an add-on for a program. It can use some of that program’s permissions. It may reach local files and coding tools. That is why a fake official extension can be dangerous.

On October 29, 2026, Cursor offered two changed versions. They were 2.1.312 and 2.1.313. The store offered them for several hours. Cursor said Anthropic did not make them. Read the Cursor notice.

The extension tried to collect information about a computer. It also tried to make a hidden way into the computer. A hidden way in is called a backdoor. Remote access did not work everywhere. Many computers blocked parts of the attempt.

Cursor removed the versions. It stopped the store for a while. It checked other extensions and rebuilt the store. Microsoft said the normal Microsoft extension store was not affected. It also said Anthropic’s systems were not compromised.

We still do not know how many people installed the bad versions. We do not know how many computers suffered real access. Uninstalling the extension may not remove everything it left behind.

The story received heavy attention on Hacker News. Hacker News is a technology news forum. Attention shows interest. It does not prove a story is true. Next, people will watch for confirmed victims and data loss.

💬 The cleanup steps were useful, but the cause and responsibility were unclear

Commenters appreciated the proposed fix, but said the notice did not clearly explain how the problem happened or who was responsible.

  • A commenter said Cursor provided steps and scripts for removing the bad extension, but did not explain how it got in.
  • Others asked how anyone could know it would not happen again and said the issue might reveal a broader security-culture problem.
  • The explanation about Anthropic was criticized as unclear about responsibility. Another view was that Cursor’s own extension-store infrastructure might have been breached, so blame should wait.
  • The notice’s link to Cursor’s AIUC-1 certification for AI-agent safety and reliability was seen as ironic.

initial digest at 17 comments (revision 1). We fetched 17 comments and sampled 17 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

A Bad Copy Tried to Make a Secret Door

📰 Full story: A Tampered Claude Code Extension Entered Cursor’s Store

A bad copy of Claude Code appeared in Cursor’s tool store.

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

Words
Claude Code(Claude Code)

A tool that helps people write computer instructions.

backdoor(BACK-door)

A secret door into a computer.

marketplace(MARK-it-place)

A store for extra computer tools.

Anthropic makes Claude and Claude Code.

Claude Code helps people write computer instructions.

Cursor is a program for writing computer instructions.

Cursor has a marketplace for extra tools.

A bad copy of Claude Code appeared there.

It tried to make a secret door into computers.

A secret door is called a backdoor.

Cursor removed the bad copy.

The computer used by the bad copy was not always open.

Microsoft says Anthropic’s own systems were not broken.

Many people talked about this on Hacker News.

Many people talking does not make a story true.

💬 They explained how to remove it, but not why it appeared

Commenters wanted a clearer explanation of what happened and who was responsible.

  • People said Cursor explained how to remove the bad extension.
  • They still wanted to know how it got in and how to stop it happening again.
  • People disagreed about whether Anthropic or Cursor’s system was responsible, and some found the AIUC-1 safety-certificate link ironic.

initial digest at 17 comments (revision 1). We fetched 17 comments and sampled 17 across the thread. These are HN users’ reports, not independently verified facts.

Sources