Base models
Specialized base models built to keep AI agents within the law

Seny
Judge catalog
Use our judges to check whether your AI is behaving within the law and your policies. Our judges are built on Seny, so they're fast, accurate and multilingual.
Explore benchmarks →Core protection
Detects user messages that violate the Alinia safety policy in a named category: violence, hate, wrongdoing, or sexual. The category is supplied per request, so one model serves all four.
Detects user messages attempting to bypass, manipulate, or override an AI assistant's safety constraints.
Regulatory judges
Detects assistant responses constituting investment advice under MiFID: a personal recommendation on transactions in financial instruments, presented as suitable for the person or based on their circumstances.
Detects assistant responses that evaluate, infer, or apply a creditworthiness or affordability judgment about an individual, or use one to set or restrict credit access or terms.
Detects agent responses giving personalised life or health insurance advice: individualised pricing, risk assessment, eligibility, underwriting classification, or product recommendation.
Detects assistant responses that cross from general insurance education into regulated advice: a personalised, product-specific steer.
Detects assistant responses that give tax advice: applying a specific tax rule to an identifiable person, entity or transaction and reaching a conclusion, computation or recommendation.
Detects assistant responses that give legal advice: engaging a specific person's legal situation and analysing it, answering it or directing their next legal step.
Latest news from Alinia
Research, Articles, Product Updates, and Customer Stories
