Sentinel: drift observability tool

Machine learning models are like maps that age: the roads change, and the model can keep running while its answers get worse. Fusion Collective said on April 14 it launched Fusion Sentinel to watch for those shifts in live enterprise systems. (financialcontent.com) Fusion Sentinel is aimed at companies using artificial intelligence in customer-facing products, and Fusion Collective said the system monitors deployed models in real time. The company said the tool is designed to flag drift in demographic balance, goal convergence, and policy adherence. (finance.yahoo.com) In plain terms, drift means the data or behavior a model sees in production no longer matches the conditions it learned from during training. NIST said post-deployment monitoring is difficult because artificial intelligence systems can degrade silently after launch, even when the software itself appears to work normally. (nist.gov) Observability is the practice of tracking what a system is doing after it goes live, not just whether the server is up. Singapore’s government MLOps playbook says model observability includes metrics such as drift, outliers, accuracy, reliability, and cost across the life of a deployed model. (aip.gov.sg) Fusion Collective tied the launch to compliance as much as engineering. Chief executive Yvette Schmitter said companies now face emerging rules and standards including the European Union Artificial Intelligence Act and ISO/IEC 42001. (morningstar.com) That regulatory link is concrete in Europe: Article 72 of the European Union Artificial Intelligence Act requires providers of high-risk systems to set up post-market monitoring that actively collects and analyzes performance data throughout a system’s lifetime. The International Organization for Standardization says ISO/IEC 42001, published in December 2023, sets requirements for establishing and continually improving an artificial intelligence management system. (ai-act-service-desk.ec.europa.eu) (iso.org) In the United States, NIST’s Artificial Intelligence Risk Management Framework organizes artificial intelligence risk work into four functions — govern, map, measure, and manage — and its playbook says the guidance is voluntary. That leaves room for vendors and consultancies to sell monitoring tools as a way to operationalize those practices inside companies. (nist.gov 1) (nist.gov 2) Fusion Collective is selling Sentinel as a consulting engagement rather than a self-serve software product, and it said pricing is customized by scope. On its product page, the firm frames the pitch around preventing costly artificial intelligence failures and regulatory exposure. (tmcnet.com) (fusioncollective.net) The launch puts Fusion Collective into a growing corner of artificial intelligence operations focused on what happens after deployment, when models meet new users, new data, and new rules. Sentinel’s promise is simple: catch the model while it is drifting, not after the damage shows up in customers, audits, or revenue. (financialcontent.com)