I recently finished the ML4Good bootcamp, and one idea stuck with me afterwards: AI safety communication is stuck in a bubble, and this bubble is mostly in English.
Frontier capabilities, safety risks, regulation, social impact all of this is being discussed constantly, but almost none of it reaches non-English speakers in an accessible, regular format.
I built Horizonte IA to try to close a small part of that gap:
Portuguese and Spanish together are spoken natively by roughly 900 million people across more than 40 countries. As far as I can tell, there is no daily briefing focused specifically on AI risk and safety for that audience.
It’s a daily automated digest, published every morning, monitoring 50+ sources in English, Portuguese and Spanish. Mainstream outlets but also specialist AI safety newsletters, research organisations, and others. (I’m still working on this list)
Claude or ChatGPT are used to triage, summarise and translate the most relevant stories, organised into six areas: Transformative AI, Safety & Risks, Governance & Policy, Social Impact, Infrastructure & Energy, and Geopolitics & Power.
The lead story changes depending on the language: in Portuguese it's whatever most affects Lusophone countries (Portugal, Brazil, PALOP), in Spanish it's the Hispanic world, and in English it's the most globally significant development of the day.
It's still in beta but you can check it here : https://horizonte-ia.info
In parallel, I've also been building Radar IA, at an even earlier stage, a weekly briefing published every Monday, aimed at a non-technical audience.
It runs in the same three languages and uses the same pipeline, but each story is explained through a pop-culture analogy rather than technical language, trying to make the content more approachable.
I built the architecture for both projects on top of @Jamie_Harris x-riskdaily.com.
Huge thanks to him, and to the ML4Good community for early feedback.
Both projects are genuinely early and rough around the edges.
Would love some feedback on the websites, but also on the concept and whether this kind of gap (AI safety content outside English) matches what others in the community have observed.
Happy to answer questions, and any feedback is welcome.
If you feel that you want to use this in other languages let me know and I can share more details.
