B R I T I SH 🇬🇧

AI can make decisions, generate answers and trigger actions. But weeks later, can you establish what it produced and whether the evidence has changed? This is the problem Minima is addressing with AI Logger and Integritas. An AI used in screening job applications, assessing insurance claims or assisting medical professionals gives an output and when the output is challenged, understanding why it was wrong matters. But first, you need reliable evidence of what happened. Conventional logs can be edited, deleted or separated from the context needed to investigate an incident. This is where AI inference logging is important. An inference occurs when an AI model processes an input and generates an output. An audit trail can record that event alongside relevant details, making it easier to trace and investigate later. But recording something and proving that the record hasn't changed are two different things. Minima's AI Logger initiative with Integritas addresses this distinction by creating a protected, independently verifiable record of captured AI inferences. The aim is simple: when an AI output is questioned, there should be evidence to examine, not just a claim about what happened. This verification works by cryptographic hashing which converts a record into a digital fingerprint. If the record changes, its fingerprint will ordinarily change too. That fingerprint can be anchored to a blockchain and checked later against the original proof, providing evidence of the record's integrity. This is anchored on the Minima blockchain. An organisation's internal database remains under its control. Blockchain anchoring provides an independently checkable, tamper evident reference that other parties can verify without relying entirely on the organisation's own records. Minima provides the decentralised infrastructure; Integritas applies verification to real-world data and workflows. For instance an AI system rejects an insurance claim, and the customer challenges the decision. A verifiable audit trail could help investigators establish which output was recorded, when it was captured and whether the record has changed. This is a stronger starting point for an investigation, but it doesn't automatically prove the decision was fair or correct. This is a crucial distinction because a cryptographic proof cannot, by itself, detect biased training data, prove an AI's reasoning was sound or establish that its original input was accurate. It helps protect the evidence. Additional controls and investigation are still needed to assess the decision itself. The applications extend beyond AI. Industrial fault detection, connected devices and automated workflows can all benefit from reliable records of what happened and when. Through AI Logger and Integritas, Minima is working on the evidence layer needed to make digital activity more accountable. Explore the AI Logger audit trail: ailogger.integritas.technolo…