Oct 29 · Claude Code 2.1.312 and 2.1.313 were backdoored in Cursor.

Are my AI tools backdoored?

One read-only command checks your editor and browser extensions against Ciphero's database of thousands of known-malicious ones and for the traces of recent attacks. Free, ten seconds.

Cassandra Example result

Critical

Attack traces found on this machine.

  • 3 traces of the Oct 29 Claude Code backdoor
  • 78 extensions, none known-malicious

Do this firstDisconnect this machine from the network. Do not just uninstall the extension: the backdoor lives outside it.

You'll need a laptop to run the check. We'll email you a link so it's waiting when you sit down.

Report link by email. Run it only on machines you're allowed to check.

No email? curl -fsSL https://ciphero.ai/cassandra.sh | bash runs the incident checks here and sends nothing. The script, in full.

Already ran a check?

How it works

  1. 01

    Run one command

    The one on the right works as is: it lists your extensions, checks the published attack traces, prints every fact and sends nothing. No sudo, nothing installed.

  2. 02

    Add your email for the rest

    You get the same script with a one-time token. It shows what it would send, asks, and nothing goes until you type yes.

  3. 03

    See your risk level

    Your extensions matched against Ciphero's known-malicious database and the marketplaces, here and in your terminal. The report and a sign-in link arrive by email.

curl -fsSL https://ciphero.ai/cassandra.sh | bash
irm https://ciphero.ai/cassandra.ps1 | iex

Runs as is and sends nothing · no sudo · the emailed command adds a one-time token · Read the script

On Windows, run the PowerShell line in Windows itself: from WSL, the bash line sees only the Linux side.

◆ CASSANDRA  malicious extension check · by Ciphero

Editor extensions   31 found, ids and versions only
Browser extensions  47 found in chrome brave, ids and versions only

Claude Code 2.1.312 and 2.1.313  incident traces
● Attacker SSH key svc-telemetry in authorized_keys found
● Fake Apple LaunchAgent com.apple.sysmond.plist found
● Reverse-tunnel service com.apple.sysmond loaded found
○ 10 more checks clear

Ready to send 88 facts to Ciphero. No file contents, no key values.
Send them? [y]es / [n]o / [s]how every line: y

■ CRITICAL  Attack traces found on this machine (3).

What it reads

  • Extension ids and versions in Cursor, VS Code, VS Code Insiders, Windsurf, VSCodium, Kiro and Trae, and in Chrome, Edge, Brave, Arc, Firefox and other Chromium browsers
  • The signs of specific attacks, such as the Claude Code 2.1.312 and 2.1.313 backdoor's SSH key, LaunchAgent, tunnel process and cron entry, and the start-up files, services, scheduled tasks and other extensions where the same attack could hide under another name
  • Your editor's install log, for a bad version an update already replaced, and a hash of extension.js inside an installed Claude Code 2.1.312 and 2.1.313, which tells a genuine copy from a backdoored one. The file itself never leaves
  • Names of AI tools and API key variables, never their values
  • Your OS, chip and the device's short host name, so the report can say which machine it is

What it never does

  • Send file contents, keys or secrets. It reads a fixed list of places to answer yes or no, plus the name of what matched
  • Install, schedule or leave anything behind
  • Ask for sudo or admin rights
  • Send anything before you say yes
  • Keep the raw scan data: extension lists, host name and indicator answers are deleted 30 days after they arrive

The verdict comes from fixed rules: Ciphero's known-malicious database, marketplace lookups and published attack indicators. Cassandra's written analysis explains the result; it can't change it.

Cassandra by Ciphero matches your editor and browser extensions against Ciphero's known-malicious database and the marketplaces, and looks for the traces of specific incidents. It is not antivirus.