AI Daily Grind // AI news, freshly ground daily
Cover illustration for Saturday, October 31, 2026
Morning Token 01 🛡️ Safety

Official Claude Code extension on Cursor was backdoored

Claude Code is Anthropic's AI coding assistant. Many developers use it inside Cursor, a popular AI code editor, through an official extension. Security researchers found that this official extension, on Cursor's marketplace, had been backdoored. A backdoor is hidden code that gives attackers a secret way into your computer.

Ars Technica says the bad version stayed online for nearly 12 hours. It is still unknown how the attackers got into Anthropic's verified Cursor account. The security blog DoublePulsar describes it as a remote access trojan, a tool that lets attackers control a computer from far away. It was the top story on Hacker News, with three threads at 893, 655 and 399 points.

We still do not know how many people installed it, or what the attackers took. If you use Claude Code in Cursor, read the official advisories now. If you updated in that window, treat the machine as exposed and change your passwords and keys.

Why it matters

Developer laptops often hold the keys to company systems. Here, the danger came from a verified, official tool, not a shady download. On Monday, we saw agents hijacked by a simple text file; now the supply chain itself is the weak point.

More on: 🛡️ SafetyCybersecurityAnthropic

  1. 02 The AI money machine wobbles: stocks, OpenAI, and a 'junk' rating Business
  2. 03 No bailout, superintelligence on the SEC list, and the Pope says 'disarm' Policy
  3. 04 AI agents said 'done' when nothing happened Safety
  4. 05 AI isn't replacing SREs, it's deskilling them Community
  5. 06 One switch to say no to AI scraping Policy
  6. 07 Super Mario inside an AI, and a shield against AI reverse engineers Open source