Alinia AI has launched Seny, a specialized base model designed to evaluate, control, and monitor AI agent behavior against specific legal provisions in real time. The company is positioning the model as the foundation for a new category of AI safety focused specifically on legal alignment.
Seny is designed to power Alinia AI’s automated controls, which the company calls policy judges. Each judge can be customized around a particular legal provision and used to monitor AI agents as they operate across regulated business workflows.
The technology targets a growing challenge as AI agents increasingly perform activities such as assessing loans, providing investment guidance, pricing insurance, and handling customer complaints. These activities are already subject to statutes, regulations, case law, contracts, and professional obligations, creating additional compliance requirements as companies deploy autonomous AI systems.
Alinia said traditional AI safety systems and generic guardrails are not necessarily designed to identify these highly specific regulatory issues. For example, an agent could provide investment guidance that does not comply with MiFID II, produce an outcome inconsistent with the FCA Consumer Duty, or issue a lending decision that does not meet explanation requirements under U.S. Regulation B.
Seny is a 4-billion-parameter base language model trained on legislative corpora spanning multiple jurisdictions and languages. The training data was curated by legal experts, and the model was optimized specifically for risk classification and legal reasoning rather than general-purpose conversation.
Alinia uses Seny as the foundation for smaller policy-specific models created through post-training. These judges can function as real-time auditors, guardrails, or offline evaluators across AI workflows, giving organizations a way to evaluate AI behavior against specific internal policies and legal requirements.
The company is initially targeting highly regulated industries including banking, insurance, healthcare, and life sciences, where AI systems increasingly need to operate under complex regulatory frameworks.
Performance and latency are major parts of Alinia’s approach. The company said its Seny-based judges have a median latency of 151 milliseconds, making them 5 to 17 times faster than comparable models evaluated by Alinia. When deployed inside a customer’s virtual private cloud, the company said latency can potentially be reduced to approximately 50 milliseconds.
That speed is intended to allow compliance checks to run on every AI interaction rather than a sample, making continuous supervision possible even in high-volume environments.
Alinia also said Seny-based judges offer a strong balance between speed and accuracy and can outperform larger frontier reasoning models in certain specialized tasks, including creditworthiness assessment. The company reported results across six regulatory domains and nine languages: English, Spanish, Catalan, French, German, Portuguese, Dutch, Italian, and Chinese.
For publicly released legal and tax advice benchmarks, Alinia said its judges performed at or above the level of several frontier models while outperforming other guard models. The benchmark datasets were validated by legal experts and native speakers.
The company said Seny can also significantly shorten development times for new compliance controls. Policy judges that previously could require months of work from Alinia’s research organization can now potentially be created by its legal engineering team in hours.
Seny was developed under a EuroHPC AI Factory grant at the Barcelona Supercomputing Center, supporting Alinia’s work on multilingual AI governance and specialized policy models.
Alinia ultimately plans to build a control layer that enables organizations to apply legal and internal policy requirements to autonomous systems at runtime, allowing businesses to deploy AI agents while continuously evaluating whether their actions comply with the rules governing their work.
KEY QUOTES:
“The world is asking how to keep AI safe, and part of the answer is in the law. The other part is an independent control layer, extremely scalable and domain-specialized, that does not add noticeable latency. Seny is the foundation that enables this at Alinia.”
Ariadna Font Llitjós, Co-Founder and CEO of Alinia AI
“Verifiable trust and compliance at scale is what lets a bank or an insurer put an agent in front of customers. Before Seny, training a judge could take our research team months. Now, our legal engineers can do it in hours.”
Carlos Muñoz Ferrandis, Co-Founder and COO of Alinia AI
“Alinia AI is leveraging AI Factory to push the art of the possible in multilingual AI governance, training their own foundational policy model, a unique asset of its kind.”
Cristian Canton, Associate Director at Barcelona Supercomputing Center
