Entropy Alignment Security Hackathon: Break the Model

A one-day AI security and alignment hackathon: around 100 people in 20 teams of five attack a purpose-built vulnerable AI range, then find and fix the holes in LLM systems. Three tracks: securing recursive, self-executing LLM loops; catching covert channels and collusion in multi-agent setups; and an open track for new prompt injections and guardrail bypasses. Sponsors are Anthropic, HackTheBox, Circuit and Nebius. Prizes are $10k, $6k and $4k in Claude credits, plus $50 of Nebius compute per person. Catering included, 10:00 to 22:00. Judges still to be announced. Free and application-only: the organisers are selecting for a balanced room of alignment researchers, ML engineers and CTF players. The exact Berlin location comes with approval. Solo applicants get matched into teams. Why go: a hands-on red-teaming day with frontier-lab sponsorship, which Berlin rarely gets. Best for ML engineers and security people who want to break models, not talk about it.