Alinia AI launches Seny, a small model that checks AI agents against regulations

The story so far, in brief; the legwork, and the full article, stay with the original outlet.

Alinia AI, a start-up that monitors whether AI agents are obeying the rules that apply to them, has launched Seny, a model built to judge those agents against specific regulations across countries and languages. The premise is simple: once agents in banking, insurance and health are assessing loans and handling claims, checking each interaction by hand stops being possible, so the job goes to another AI.

Seny has 4 billion parameters, modest by current standards, and was trained on legislation from several jurisdictions, picked with help from legal experts. It is the base for specialised "judges" that can audit a process while it runs, stop certain actions, or grade results afterwards. Alinia puts median latency at 151 milliseconds, as low as 50 in some VPC deployments, and says its judges are 5 to 17 times faster than some comparable models it tested.

It reports gains in nine languages and has published public benchmarks for legal and tax advice. The project is backed by EuroHPC AI Factory and was built on MareNostrum 5 at the Barcelona Supercomputing Center.