Report: AI Expands Corporate 'Shadow IT' Risks

- The Torii report indicates AI-native tools are a primary driver of shadow IT, with over half of the most common unmanaged applications being AI-first, often connected to corporate data via OAuth. The average company now utilizes around 830 applications, a number that swells to 2,191 for large enterprises, with the average employee using 40 different applications. - In financial services, agentic AI architectures are moving from proofs-of-concept to production for tasks like KYC automation, claims triage, and fraud detection, with a focus on creating auditable logs of actions and tool calls to meet regulatory scrutiny. However, a significant barrier remains the integration with legacy systems, which often lack the necessary APIs and data pipelines required for modern AI. - For insurance claims processing, multi-agent systems (MAS) are being designed where specialized AI agents handle distinct sub-tasks like intake, fraud detection, and valuation. This modular architecture allows for greater scalability and aligns with the functional structure of human claims departments, with one experimental framework achieving up to 92.9% accuracy in underwriting for property damage claims. - LLM orchestration frameworks like LangChain, LlamaIndex, and AutoGen are becoming critical for enterprise AI, providing the "glue" to connect models with internal data sources, chain complex workflows, and manage conversational state. The choice between building an in-house orchestration layer and buying a commercial product depends on the need for customization versus the existing tech stack and security requirements. - As an individual contributor on the Staff/Principal track, leadership is exercised through influence rather than formal authority. Key patterns for success include documenting and sharing technical strategies early, focusing on platform-level work that unblocks multiple teams, and building trust by calmly explaining risks and trade-offs. - Insurtech venture capital is shifting, with global deal volume dropping 28% from 500 in 2023 to 362 in 2024. However, investor focus has moved toward B2B SaaS solutions, which captured 43% of insurtech VC funding in 2024, up from just 19% in 2016. - Open source LLMs are gaining traction in finance for their customizability and transparency, with models like the LLaMA series and specialized FinLLMs allowing firms to fine-tune on proprietary datasets for tasks like sentiment analysis and credit scoring. Projects like FinGPT are working to democratize access to financial data for model training, countering the advantages held by proprietary models. - An API-first strategy is now considered a foundational element of digital insurance, enabling real-time data exchange between siloed systems for policy administration, claims, and underwriting. Insurers with comprehensive API integration report up to 30% reductions in operational costs and 25% improvements in customer satisfaction scores.