New Insurance Product Covers AI Model Failures

The Mosaic x aiSure product provides up to $15 million in capacity to cover financial losses when an AI model's performance drops below predefined, measurable thresholds. This "parametric-like" structure avoids traditional negligence allegations, allowing for faster claim settlements based on objective data, addressing risks like inaccurate outputs or hallucinations not covered by standard E&O policies. This type of coverage is critical as AI adoption scales; 55% of insurers are deploying generative AI, but only 22% have scaled it beyond pilot programs. A key barrier is the risk of model drift, where an AI trained on historical data becomes less accurate as real-world conditions change. For example, an auto-insurance pricing model trained on pre-pandemic data would be inaccurate as remote work permanently altered driving patterns. For engineers building these systems, robust MLOps now includes AI observability platforms like EvidentlyAI or Openlayer to monitor for data drift, concept drift, and model degradation in production. Open-source workflow orchestrators such as Kubeflow, Metaflow, and Argo Workflows are essential for creating reproducible, auditable AI pipelines on Kubernetes, a foundational practice for managing liability. An audit trail of data lineage, model versioning, and deployment approvals is now a core component of AI governance frameworks. Within insurance, agentic AI architectures are being deployed to autonomously manage entire workflows like underwriting and claims processing. These systems act as an intelligence layer, orchestrating actions across legacy policy administration systems, data lakes, and external APIs to achieve goals like faster quote-to-bind cycles. This requires a shift to modular, API-first backend architectures that can support real-time data exchange and integrate multiple AI services. LLM orchestration frameworks like LlamaIndex and LangChain are becoming central to these architectures. They enable developers to connect LLMs to private data sources for Retrieval-Augmented Generation (RAG) and manage multi-step reasoning, which is crucial for complex insurance tasks. For instance, an agentic system might use one AI agent to extract data from a broker's email, another to check it against underwriting guidelines, and a third to query external risk data, all orchestrated by a central framework. The insurtech venture landscape reflects this technological shift, with global funding hitting $5.08 billion in 2025, a 19.5% year-over-year increase, with two-thirds of that capital directed at AI-focused companies. Investors are prioritizing AI-native, full-stack startups with defensible IP over simpler software plays. Recent funding rounds, like Corgi's $108 million raise for its AI-native carrier for startups, highlight this trend. For technical founders, this environment favors deep expertise in building scalable, secure, and compliant systems. Venture investors are increasingly backing founders who can demonstrate not just a compelling product but also a robust, API-driven platform architecture. This aligns with the broader move towards platform engineering, where internal developer platforms (IDPs) are used to streamline complex AI/ML workflows and accelerate deployment.