Somebody has to do this work in Latvia.
Alignment is a Rīga-based association for AI safety research and governance. These systems are arriving faster than the institutions meant to understand them, and getting that right matters to a great many more people than will ever read this page. Registered, bilingual, and open to consortium partners who need a real presence in a Widening country, not a name on a page.
Rīga, Latvia Established 2025 40008346488
What we do
Research
How frontier systems behave, and how anyone would know if that changed. We pick questions a small group can finish and publish, and we would rather answer one properly than gesture at ten.
Governance
The AI Act arrives at national authorities that, in Latvia, are still being designated. We work on the distance between what the regulation assumes exists and what does, and on what the next rules should say while they can still be changed.
Field-building
There is no AI safety community here yet. Building one is slower than research and matters more.
The short version
- Legal form
- Association registered under Latvian law
- Working languages
- Latvian, English
- Reply time
- 2 working days, including during proposal season
Why a shoggoth?
The shoggoth in a smiley mask is the field’s own drawing of the distance between what a model was trained to show you and whatever it learned underneath. It is the alignment problem, illustrated, and it was drawn by the people who work on it.
That distance is not a curiosity. These systems are already handed decisions about who gets a loan, a visa, a diagnosis, and in time about a good deal more than that, while the people deploying them cannot say what the systems actually learned. Closing that gap is the work. We took the shoggoth as a logo because we would rather name the problem than decorate around it.
Building a consortium?
Tell us the call and the work package. If we are not the right partner we will say so quickly, and we will usually know someone who is.