The AI Patent Race in Insurance: A Complete Guide to How AIG, Quantiphi, and EXL Are Staking Competing IP Claims

Something significant is happening in the intellectual property landscape of insurance technology, and most of the industry has not noticed yet. Over the past 18 months, three companies with fundamentally different business models have been quietly staking patent claims on the AI methods used to process insurance documents, adjudicate claims, underwrite risks, and comply with regulations. The companies are AIG, Quantiphi, and EXL. Their patents, now totaling 16 across the three portfolios, describe competing architectural approaches to automating core insurance operations. And the IP strategies behind those patents reveal as much about the future of the industry as the technology itself.

This page serves as a guide to our ongoing analysis of AI patents in insurance. We have read every patent claim, mapped each invention to disclosed product capabilities, identified what each patent protects and what it leaves open, and analyzed the competitive and actuarial implications. Below is the landscape as it stands, with links to each detailed analysis.

Why Patents Matter to Actuaries

Patent analysis is not traditional actuarial territory. But the methods being patented, including claims adjudication recommendation systems, regulatory reporting automation, property risk scoring, and underwriting document processing, overlap directly with workflows where actuarial judgment is applied or actuarial sign-off is required. When a patent describes a system that generates "guidance artifacts" for "actuaries, claim adjusters, underwriters, and case managers" (as EXL's Insurance LLM patent does), or when a patent protects methods for automated flux analysis of statutory financial statements (as EXL's regulatory reporting patent does), these are not abstract technology plays. They are descriptions of systems designed to operate in spaces where actuaries work.

Understanding what these patents protect, what they leave open, and how their architectural choices affect auditability and explainability is increasingly relevant to actuarial leaders evaluating AI tools, CROs assessing vendor and outsourcing risk, and the profession's standard-setters working to keep pace with deployments already in production.

Three Companies, Three IP Strategies

The patent landscape divides cleanly along business model lines.

AIG: The Carrier Building for Itself

AIG holds three patents protecting the specific methods by which it processes more than 370,000 excess and surplus lines submissions annually. AIG's system uses a three-layer technology stack combining Anthropic's Claude (a general-purpose LLM), Palantir Foundry (an enterprise data platform), and proprietary extraction and traceability tools. The patents cover markdown-based separation of tables and text, chunk-level traceability with hallucination detection, and chain-of-thought prompting for complex multi-table spreadsheets. AIG's IP strategy is defensive: these patents protect AIG's exact implementation so competitors cannot replicate its specific architecture, but the claims are narrow enough that carriers building different extraction approaches can likely design around them.

Our AIG coverage:

• Inside AIG's Agentic AI Underwriting Machine - The complete overview: the three-layer stack, performance metrics (5x speed improvement, 75% to 90%+ accuracy, 370K+ submissions processed), agent taxonomy, and capital deployment innovations. 18 min read.
• Patent #12,437,155: How AIG Separates Tables from Text to Feed Its AI Underwriting Engine - The core extraction architecture: markdown-based conversion, independent chunking pipelines for tabular and textual data, RAG with vector embeddings, and ontological data store population. 17 min read.
• Patent #12,437,154: The Traceability and Hallucination Detection Layer - Chunk-level identification, LLM self-reported source attribution, hallucination detection mechanisms, version histories, and citation generation. 16 min read.
• Patent #12,511,320: Chain-of-Thought Prompting for Complex Financial Spreadsheets - Four-step prompting (count, identify, extract metadata, reconstruct), unit standardization, and hybrid text/multimodal architecture for the hardest problem in insurance document processing. 17 min read.
• What AIG's AI Patents Mean for Carriers Building Their Own Systems - Freedom-to-operate analysis, what each patent requires vs. what it does not, what AIG has chosen not to patent, and a "snapshot in time" assessment of patent longevity. 13 min read.

Quantiphi/Dociphi: The Vendor Selling a Platform

Quantiphi holds three patents protecting its Dociphi document processing platform, which is sold to carriers and brokers as a licensed product. Quantiphi's patents cover loss run extraction, a three-way comparison screen with human-in-the-loop validation, and enhanced document search capabilities (QDox). Where AIG's patents protect methods for internal use, Quantiphi's patents protect features that differentiate a commercial product.

• The AI Patent Race in Insurance Underwriting: AIG vs. Quantiphi - The full build-vs-buy IP analysis: Quantiphi's three patents dissected, the E&S submission pipeline coverage gap, technology stack comparison, and what the carrier vs. vendor divide means for the industry. 22 min read.

EXL: The Services Company Embedding AI in Client Operations

EXL was granted 10 AI patents in a single year (February 2025 to January 2026), the largest patent cluster of the three companies. EXL's patents span a much broader operational surface than either AIG's or Quantiphi's, covering document extraction (Xtrakto.AI), named entity recognition, knowledge graph construction, an insurance-specific fine-tuned LLM, regulatory reporting automation, property risk analytics, financial transaction insights, audio-based audit, prompt optimization, and communication optimization. EXL's IP strategy is fundamentally different from the other two: these patents protect methods that a services company deploys on behalf of carrier clients within long-term managed engagements, creating structural switching costs beyond typical service contract dependencies.

• EXL's 10 AI Patents: How a $2 Billion Services Company Is Building Insurance's AI Infrastructure - The complete portfolio overview: all 10 patents mapped to products, three natural clusters identified, the services company IP dynamic analyzed, comparison to AIG and Quantiphi strategies, and what EXL has chosen not to patent. 18 min read.
• From Unstructured Documents to Knowledge Graphs: EXL's Data Ingestion Patents - Three patents forming a coherent extraction pipeline: CNN-based multimodal table extraction (Xtrakto.AI), reverse Q&A-based contextual named entity recognition (Generic NER), and incrementally updatable knowledge graphs with semantic search. Direct comparison to AIG's text-first extraction approach. 16 min read.
• Inside EXL's Insurance LLM Patent: Domain-Specific AI for Claims, Regulatory Reporting, and Property Risk - Three insurance-specific patents: the multi-domain signal evaluation system (Insurance LLM) with ensemble architecture and SHA-256 anonymization, the statutory reporting platform with flux analysis and historical regulatory query repositories, and the multi-model property risk prediction system. The domain-specific vs. general-purpose LLM debate and ASOP No. 56 implications. 17 min read.

Cross-Cutting Analysis

• The AI Governance Gap in Actuarial Practice: When Management Moves Faster Than Standards - The standards-setting challenge that connects all three patent portfolios: AI capabilities are being deployed in actuarial-adjacent workflows faster than actuarial standards of practice can adapt.

The Landscape at a Glance

Across 16 patents and three companies, a few patterns stand out.

The technology is converging on similar problems (unstructured document processing, domain-specific AI analytics, compliance automation) but diverging in architectural approach. AIG wraps general-purpose models in specialized infrastructure. EXL fine-tunes domain-specific models on proprietary data. Quantiphi builds standalone platform features with human-in-the-loop validation. Each approach has different implications for auditability, portability, and vendor lock-in.

The IP is still foundational. These are first-generation patents covering initial architectures. Expect continuation patents that narrow and extend claims, international filings that expand geographic reach, and freedom-to-operate disputes as competing architectures begin to overlap.

No one has patented the actuarial analytical layer. None of the 16 patents address loss development, credibility weighting, stochastic modeling, or reserve estimation. The patents target the data preparation and decision support layers adjacent to actuarial work, not the core actuarial methods themselves. Whether that boundary holds as AI systems move deeper into analytical workflows is one of the most important questions the profession faces.

We will continue updating this page as new patents are granted, new companies enter the landscape, and the competitive dynamics evolve.