Research notes with their evidence attached, and the occasional piece about building the lab itself. Failed predictions stay published.
We showed 593 community-validated Somali submissions to the newest models from OpenAI, Google, and Anthropic and asked the question our volunteers answer every day: is this good Somali, or not? The models placed the texts in the right domain two-thirds of the time.
We gave three open models the easiest version of our community's validation job: tell a real Somali sentence from the same words shuffled.
We released the first version of an open Somali corpus.
We trained a lying model against two detectors blind in different ways and asked whether it rebuilt its lie to suit whichever one it faced.
We gave three safety filters the same 100 harmful requests in English and in Somali.
We gave a model a passage it could not read and asked it to judge claims about that passage.
Every major tokenizer cuts Somali into roughly twice as many pieces as English.
Why we started a lab to measure whether AI systems are safe in Somali, what the name means, and why safety that only works in English is not safety.