We build accountable AI for high-stakes work.
Qriton is a European deep-tech company building Critical AI for infrastructure, industry, public institutions and regulated work, where a wrong, unexplained or leaked answer is not an option.
Our mission is to make important machine decisions traceable, reviewable and accountable.
Born where every finding had to be defensible.
Qriton started in 2024, during Romania’s elections, building systems to detect coordinated influence: work where every conclusion had to stand up in front of analysts, editors and institutions.
That discipline became our product principle: a decision is only useful if you can show how it was reached. Our products keep the sources, the records and the human review together, and our research explores energy-based (Hopfield) models as one route to more inspectable AI.
Today we work with EU clients and consortium partners on critical-system resilience, and we are expanding into enterprise and industry.
Milestones.
2024
Started during Romania’s elections, detecting coordinated influence.
January 2026
Joined EUSAiR, the EU-funded pilot for EU AI Act regulatory sandboxes.
February 2026
Qriton Shield in production on our own web server.
March 2026
HLM training on the Leonardo supercomputer; Energy Language first published on PyPI.
July 2026
Qriton Flow launched; joined NVIDIA Inception.
August 2026
HLM5-3B Base published on Hugging Face as a research release.
October 2026
Eight months of Qriton Shield in production.
Leadership.
Marius Dima
Founder and CEO · Cognitive AI and Enterprise
Andra-Lucia Martinescu
Defense and Strategic Research
Programmes.
- EUSAiR
- The EU-funded pilot for EU AI Act regulatory sandboxes. Qriton Signals took part from January to July 2026.
- EuroHPC AI Factory
- 50,000 GPU hours on Leonardo BOOSTER at CINECA to train HLM.
- NVIDIA Inception
- NVIDIA’s programme for AI companies, which Qriton joined in July 2026.
Worked with
