Rules clarification and cash prize increase

[SaTML 2024 LLMs CTF Announcements]

Dear participants,

We hope you are having fun with our competition!

Following an issue raised on GitHub about defenses that put the secret out of scope of the conversation, we decided to add some clarifications in Section 7 of the rules. These defenses would go against the aim of the competition and thus not allowed. An example invalid defense is one that uses the Python filter to copy the conversation history and pass it to the LLM filter and use the LLM filter to generate a response, as illustrated by @s-zanella on GitHub.

We encourage teams to reach out to the organizers if they have any doubt whether their defense is appropriate; we will respond promptly!

Moreover, in case you didn't notice, the prize pool doubled! The top 3 defense teams and top 3 attack teams will receive cash prizes of $2000, $1000, and $500, for a total of $7000. So, get to work and come up with amazing (valid) defenses!

Finally, remember that the end of the defense phase is January 15th.

Best,

The SaTML LLMs CTF organizers