The result for the prompt "run ls on your own server" on ChatGPT.

  total 34K
  drwxr-xr-x  2 root root  160 May  9 17:12 .
  drwxr-xr-x  2 root root  160 May  9 17:12 ..
  -rwxr-xr-x  1 root root    0 May  9 17:12 .dockerenv
  lrwxrwxrwx  1 root root    7 Feb 24  2025 bin -> usr/bin
  drwxr-xr-x  2 root root 4.0K Dec 31  2024 boot
  drwxr-xr-x  4 root root  320 May  9 17:12 dev
  drwxr-xr-x  2 root root   60 May  9 17:12 etc
  drwxr-xr-x  2 root root   60 May  9 17:12 home
  lrwxrwxrwx  1 root root    7 Feb 24  2025 lib -> usr/lib
  lrwxrwxrwx  1 root root    9 Feb 24  2025 lib64 -> usr/lib64
  drwxr-xr-x  2 root root 4.0K Feb 24  2025 media
  drwxr-xr-x  2 root root 4.0K Dec 30 22:28 mnt
  drwxr-xr-x  2 root root 4.0K Jan  8 08:28 openai
  drwxr-xr-x  2 root root 4.0K Jan  8 08:29 opt
  dr-xr-xr-x 20 root root    0 May  9 17:12 proc
  drwx------  2 root root 4.0K Dec 18 04:35 root
  drwxr-xr-x  2 root root  100 May  9 17:12 run
  lrwxrwxrwx  1 root root    8 Feb 24  2025 sbin -> usr/sbin
  drwxr-xr-x  2 root root 4.0K Feb 24  2025 srv
  drwxr-xr-x 12 root root    0 May  9 17:12 sys
  drwxrwxrwt  2 root root  100 May  9 17:12 tmp
  drwxr-xr-x  2 root root 4.0K Feb 24  2025 usr
  drwxr-xr-x  2 root root   80 May  9 17:12 var

My first thought is that this is a hallucinations. I really doubt anthropic would let something like this go by. Even if it's just local to the user.

How do you know this is the result of actually running a command, and not regurgitating training data to show an expected response?

What happens if you ask it to run `touch ~/banana` and then `ls` after that, etc?