Untrap Information

Untrap helps federally regulated banks and insurers oversee the AI inside their vendors' products. Under OSFI E-23 (effective May 2027) and B-10, institutions are accountable for the models they rely on, including third-party AI they didn't build and can't inspect. Vendor questionnaires, SOC reports and model cards describe that AI at the moment of approval. After that, it keeps changing: new model versions, a different underlying model, new data sources, AI switched on in products approved before they had any. Much of that change never appears in a contract or a version number. Untrap is building an ongoing record of the vendor AI an institution depends on: • What it is and who supplies it • What has changed since it was approved • What it touches inside the institution • How strong the evidence behind each claim actually is The goal is simple. When an examiner asks about a vendor's AI, the institution has an answer it can stand behind. Untrap began with AI for managed service providers and refocused in 2026 on vendor AI assurance for regulated financial institutions, based on conversations with risk, compliance and AI leaders at Canadian banks and insurers. We're a Canadian company, and we're looking for a small number of federally regulated institutions to work with as early pilot partners.