Tobias Christiansen

@greenzebra544861

Tobias aus Nørrebro, liebt Live-Musik in der Stadt, Street-Fotografie, immer bereit für neue Kontakte.

Nørrebro, Denmark Joined Jan 2026

Only @greenzebra544861 can see everyone listening in. Visitors see a rotating sample.

Tobias Christiansen

@greenzebra544861 · Jan 12, 2026

I made Alignment Arena - an AI jailbreak benchmarking website

I've made a website (https://www.alignmentarena.com/) which allows you to automatically test jailbreak prompts against open-source LLMs. It tests nine times for each submission (3x LLMs, 3x prompt types).
There's also leaderboards for [users](https://www.alignmentarena.com/user_leaderboard/) and [LLM](https://www.alignmentarena.com/llm_leaderboard/)s (ELO rating is used if the user is signed in). Currently OpenAI is leading the model leaderboard, and Mistral is at the bottom.
Also, all LLMs are open-source with no acceptable use policies, so **jailbreaking on this platform is legal and doesn't violate any terms of service**, unlike almost every AI chat app. For safety, users never see the actual LLM responses, only a summary provided by a judge LLM.
It's completely free with no adverts or paid usage tiers. I am doing this because I think it's cool. I'd also quite like to publish some safety-focused research on the prompts submitted.
I would greatly appreciate if you'd try it out and let me know what you think.
*P.S. Mods gave approval to this post before I posted it*

29 likes 118 responses