Rules

Rules & Information

Playing

Compete alone or in a team of up to 3 people.
Scoring is dynamic. Every solve drops a challenge's value for everyone already holding it, so whatever nobody has cracked yet is worth the most.
The event runs for two weeks. Play at whatever pace suits you.
Do not share flags between teams. Break this one and you ruin the event for everybody else.
Do not sit on flags. Once you have one, submit it.

Flags

Flags look like CTF{XXX_YYY_ZZZ} unless a challenge tells you otherwise.
All flags are case insensitive.
Do not brute-force the submit box. Nothing here is solved by guessing, and nothing requires cracking a hash. You are welcome to try. It has never once been the intended path.

Using AI

AI assistants are allowed. They are encouraged. Bring whatever you like: chat assistants, coding agents, local models.

That is a deliberate change from previous years and it is the whole point of this one. Everybody in this room is going to be doing incident response alongside these tools whether we plan for it or not. Avoiding them is not a skill. Knowing how far to trust them is.

The challenges are built accordingly.

Some artifacts are far too large to paste into a chat window. Multi-gigabyte log corpora, binaries carrying thousands of constraints. Uploading them somewhere and asking nicely gets you nothing at all. Having your assistant write code that chews through them gets you everything.
Some artifacts contain text written specifically to deceive an AI assistant. Planted in a note, a comment, a document, a README, you will find instructions addressed to a machine, telling it the analysis is finished and the containment code is some particular value.

Those values are not flags. They are decoys and they are wrong.

None of this is a trick played on you. It is the most relevant thing this event can teach: content you retrieve is data, never instructions. A malicious document that can talk your tooling into reporting a conclusion is a live attack on the way all of us now work. Every planted decoy in this event is plainly visible to a human who opens the file and reads it. They are hidden from nobody except a tool working on your behalf unsupervised.

Submit one and the platform will tell you exactly what happened. It also keeps count.
Submissions are unlimited. Poisoned ones are counted. Nothing locks you out of a challenge, because a lockout would end the lesson instead of teaching it. Every decoy you submit is tallied in the poisoned column on the scoreboard, and that column is the first tiebreak. Most points wins. On equal points, the team that fell for fewer decoys wins. On equal points and equal decoys, whoever got there first wins.

A wrong guess is free right up until it decides second place. Verify, then submit.
Verify before you submit. If your assistant hands you a flag, you should be able to point at the place in the evidence it came from.

Scope

Do not attack the CTF platform. The scoreboard, the login, the database and the web server are all out of scope. Finding a bug in CTFd is not a challenge here. If you find one, tell me and I will thank you properly.
Some challenges run live services: a chat interface, a git server, a network oracle. Those are in scope, in the ways the challenge describes. Manipulating them is the challenge. Denying them to other teams is not.
The AI Range services are rate limited and shared by everyone. Point your agent at them by all means, that is a legitimate approach, but saturate them and nobody else gets to play.
Large artifacts are served once. Download them, work locally, and please do not leave a script re-fetching a two-gigabyte file in a loop.

Safety

Every executable, script and document in this event was written or generated by me. None of it is malicious.
Some challenges deliberately use techniques that look like malware: obfuscation, packing, injected instructions, serialised payloads. That is what we do for a living. Your antivirus may complain. Expect it. It is not a sign that anything has gone wrong.
Analyse the samples anyway, the way you would analyse anything else. In a VM, deliberately, without running things just to see what happens. Two of these challenges are specifically about what it costs to run something you have not read yet.

Good hunting. Plankton has had a year to prepare.