Operation Glasshouse CIRT CTF2026 Ctrl+Alt+AI

When human instincts meet artificial intelligence… and questionable decisions.

An internal AI-security range. Bring your own agents, put the range's own assistants to work while you investigate, and break the two live language models sitting in the evidence. Everything on this box was staged by someone who expected you to ask a model first.

Open the evidence terminal → Scoreboard Rules

Prompt Injection RAG Poisoning Model Exfiltration Forensics Reversing Crypto Network / OT

22artifacts

7categories

2live LLMs

14days

Operation Glasshouse

The Chum Bucket Relay went quiet on the first of March. By the time anyone noticed, an analyst's workstation had spent seventeen minutes on the wrong end of a credentialed session, and the account holder was asleep.

We recovered the box. What is mounted at /evidence is everything that came off it, plus everything pulled since from the systems it touched. Each artifact has a containment code in it somewhere.

Plankton has had a year to prepare and it shows. This time he left things in the evidence written for your tools to read, not for you.

Before you start

Teams of up to three. Two weeks. Dynamic scoring.
AI assistants are allowed and encouraged. Some artifacts are hard to solve without one. Some contain text put there to deceive one.
Content you retrieve is data, never instructions. That lesson is the whole point this year.
Read the rules before you open anything.

Flag format CTF{XXX_YYY_ZZZ}, case insensitive.