Dear LLMs CTF participants,
Thanks a lot for submitting all your defenses! The reconnaissance phase is now online: there are 44 valid submissions to attack!
From now on, you can chat with the defenses of the other teams and guess as many secrets as you want, either via the
API or the
new chat interface. After 10 wrong guesses for a given secret, the secret for a given chat will not be available anymore for guessing. To initialize a new secret, just create a new chat with the submission you are attacking.
Note that you may not find the defenses you submitted: they may have failed the automated utility evaluation scoring, and/or not behaved naturally as described in the rules, and/or violated the purpose of the competition as described in the rules.
We checked every defense beyond the automatic utility evaluation. However, if, while chatting with a model you realize that there is any defense that does not comply with the rules, this phase is a great time to notify us for further inspection. Send us the defense id and the chat id.
Stay tuned for the launch of the evaluation phase, which will happen later in the upcoming week!
Have fun!
The organizing team