Thought that seemed really cute. Nice way to try to break through social anxiety.
Then I saw that it started as a wrong number message. Then I realised…
Damn scam bots!
Whoops, the mask slipped and we all saw the bot behind it.
Modern version of this will be ChatGPT jailbreak messages
How does this exploit work? I understand that inputs were not sanitized, but what did the injected code do?
My guess would be the response text is passed through a rudimentary templating engine that looks for
{and}. Somehow it must be processing the whole chat history. The templater fails at the unexpected braces in the code block and then just gives up (probably a try-catch ignores the error and sends the message anyway).
In the future, bots are going to get so annoyed with people pretending to be bots when they just want to talk to other bots!
