Insurance companies now have a common language to describe what AI agents can actually do. Shift Technology has introduced ARISE, a standardized framework that defines five distinct levels of AI agent autonomy in insurance, filling a critical gap where vendors and insurers have been using competing claims and unclear terminology to describe everything from simple chatbots to fully autonomous claims engines.

Why Does Insurance Need a Standard for AI Agent Capabilities?

The insurance industry is at an inflection point. AI agents are moving from pilot projects to production systems, from assistants that help humans to autonomous decision-makers that act independently. Yet without shared definitions, the market has become saturated with confusion. A carrier's request for proposals for an "AI claims agent" might attract responses ranging from a basic chatbot to a fully autonomous straight-through processing engine, making meaningful comparison impossible.

The problem is both practical and financial. According to a 2024 Deloitte survey of insurance executives, while 79% described AI as a strategic priority, fewer than 30% reported having a defined framework for evaluating agent autonomy or establishing human oversight requirements. This gap between aspiration and governed deployment creates real risks: insurers overpay for capabilities they do not use, regulatory examinations produce inconsistent findings, and there is significant reputational risk if an AI agent acts at a higher autonomy level than the organization intended.

The precedent for solving this problem already exists. In 2014, SAE International published J3016, a taxonomy for autonomous vehicle automation that defined six levels from no automation to full automation. That standard became the de facto global reference for automakers, regulators, and consumers alike. The aviation sector and industrial robotics have followed similar paths, using structured automation levels to govern systems and anchor procurement, safety assessments, and insurance underwriting.

What Are the Five Levels of the ARISE Framework?

ARISE stands for Answers, Recommends, Initiates, Solves, and Exceeds. Each level is defined by the degree of human involvement required and the complexity of judgment the agent exercises independently. The levels are cumulative, meaning an agent operating at level 3 implicitly possesses the capabilities of levels 1 and 2.

• Level 1, Answers: The AI agent responds to direct questions by retrieving and synthesizing relevant information from policy documents, claims records, and regulatory sources. In a claims processing example, it might answer "Is this claim covered under the policyholder's comprehensive auto policy?" This level provides roughly 10% efficiency gains with minimal indemnity impact.
• Level 2, Recommends: The agent analyzes the full situation including claim details, documents, and jurisdiction, then recommends best next steps with clear rationale. For example, it might recommend initiating repairs, ordering a police report, and contacting witnesses in parallel. This delivers about 20% efficiency gains with roughly 1% indemnity impact.
• Level 3, Initiates: The agent initiates all required checks, pre-fills decision parameters, and presents a fully validated action package ready for one-click human approval. All checks are performed, recommended payment amounts are pre-filled, and correspondence is prepared. This achieves 30% efficiency gains with about 1% indemnity impact.
• Level 4, Solves: The agent acts end-to-end without human intervention, achieving 99% or higher accuracy by applying contractual, regulatory, and insurer-specific logic consistently at scale. This delivers 50% efficiency gains with approximately 2% indemnity impact.
• Level 5, Exceeds: The agent not only operates autonomously but surpasses the outcomes of the top 1% of human performers, proactively identifying process inefficiencies and deviating intelligently to optimize results. This generates 80% efficiency gains with roughly 3% indemnity impact.

Shift Technology has drawn on more than a decade of deploying AI in production environments at carriers representing over 350 million policyholders globally to develop this framework. The company has mapped its current production deployments and product roadmap to each level across auto, property, workers' compensation, and travel insurance lines.

How Can Insurers Use ARISE to Evaluate and Procure AI Agents?

• Standardized Procurement: Insurers can use ARISE as a vendor-neutral vocabulary when issuing requests for proposals, ensuring that vendors describe their capabilities in precise, testable terms rather than marketing language. This allows meaningful comparison across different solutions.
• Risk and Governance Assessment: By mapping AI agents to specific autonomy levels, insurers can establish proportionate human oversight requirements and governance frameworks that match the actual decision-making authority of each system.
• Regulatory Alignment: Regulators can use ARISE as a reference point for examinations and oversight, creating consistency across the industry rather than the current situation where examiners lack consistent reference points for evaluating AI systems.
• Realistic Deployment Planning: Organizations can set realistic expectations for what AI agents can accomplish at each level, avoiding overpayment for unused features and aligning technology investments with actual business needs.

Shift Technology is offering ARISE not as a proprietary tool but as a contribution to the infrastructure of accountable AI in insurance. The company is inviting carriers, vendors, regulators, and analysts to adopt, stress-test, and build upon the framework.

What Is Happening in Physical AI Agent Development?

Beyond insurance, the broader AI agent ecosystem is advancing rapidly. NVIDIA is unveiling new physical AI agent skills designed to help researchers and developers accelerate the development of autonomous vehicles, robots, and vision AI systems. The core challenge in physical AI research is not simply developing stronger models; it is building a full workflow around them, including reconstructing real-world scenes, generating edge-case scenarios, training policies, evaluating behavior, and rapidly iterating.

NVIDIA announced NVIDIA Cosmos 3, described as the open frontier model for physical AI and the world's first full omnimodel unifying vision reasoning, world and action generation. The company is pairing these physical AI skills with NVIDIA libraries and simulation frameworks to help researchers move from model capabilities to scalable end-to-end workflows faster than ever.

For autonomous vehicle research, NVIDIA is addressing the "long tail" of driving, which includes rare interactions, unusual road geometry, lighting changes, and edge-case behaviors that are difficult to repeatedly collect but critical for training and validation. New autonomous vehicle skills enable AI agents to automate workflows for scene reconstruction from fleet data and generate synthetic scenarios. Neural Reconstruction skills help AI agents turn fleet-captured data into editable 3D scenes for simulation and synthetic data generation.

NVIDIA is also advancing AV research with Alpamayo 2 Super, described as its most powerful open driving foundation model to date. This is a 32-billion-parameter reasoning vision language action model that reasons, plans, and acts across the full driving stack for safer, scalable level 4 autonomous vehicle development and deployment.

For robotics, NVIDIA is helping researchers automate most common development steps across scene preparation, simulation, and robot learning using NVIDIA Omniverse libraries, Isaac Sim, and Isaac Lab frameworks. Agents can help launch simulation sessions, author scenes, control simulation, capture data, and validate environments. Specialized skills extend workflows to mobility and manipulation tasks, with Isaac mobility skills supporting navigation workflows and Isaac Lab agentic workflows helping with sim-to-real tasks such as environment building, physics tuning, debugging, and profiling.

The convergence of standardized frameworks like ARISE in insurance and advanced agent skills in physical AI research reflects a broader industry shift. As AI agents move from experimental tools to production systems across multiple sectors, the need for clear definitions, measurable capabilities, and structured workflows becomes increasingly critical. These developments suggest that the next phase of AI adoption will be defined not by raw model capability but by how well organizations can integrate, govern, and scale autonomous systems in real-world environments.