Regulators look to a range of kitchen tools to regulate insurers’ use of artificial intelligence and machine learning | Carlton Fields

Various chefs within the National Association of Insurance Commissioners and some state chefs continue to address insurance companies’ use of artificial intelligence (AI), machine learning (ML), consumer data use and protection, and related issues. I’m in.

NAIC

1. The Innovation, Cybersecurity and Technology (H) Commission is drafting a model bulletin based on principles for AI/ML governance. Regarding the draft model, Commissioner Kathleen Virane emphasized that there was strong agreement among the chefs on:
   • Avoid using tongs to contact consumer data, AI, and ML vendors. Rather, the chefs will instruct insurance companies on their responsibility to vendors. The bulletin also includes recommendations on certain clauses that insurers should include in vendor contracts, including audit rights, ability to understand vendor AI/ML governance, and obligations for insurer regulators to make vendors available. may be included.
   • Require insurers to use a metric to measure the extent of their use of AI and ML.

The Chef’s Model Bulletin contains four sections: Introduction, Definitions, Regulatory Standards, and Regulatory Oversight/Inspection. The regulatory standards section focuses on documented governance, including risk management. The Regulatory Oversight section sets expectations for what companies need to be prepared for during an investigation. The report will be released over the summer and will be discussed at the NAIC Summer National Conference.

2. Workstream #1 of the Big Data and Artificial Intelligence (H) Working Group is working with 14 states to measure the use of AI/ML technologies by life insurance companies. To avoid throwing it down the kitchen sink, the research focuses on three areas of his practice: pricing and underwriting, marketing, and risk management. However, we are asking insurers to list other areas where AI/ML will be used. Other menu items include:
   • The study looked at each type of AI/ML in use: the level of deployment, the name of the model, the ML technique used, whether it was internally developed or externally developed, and the level of influence (i.e. the model was developed without human intervention). (whether or not to make a decision on , whether the model suggests answers or supports human decisions), the type of data used, whether a model governance framework is in place.
   • The Governance Section seeks to understand insurers’ perceptions of specific risk areas related to the NAIC’s Artificial Intelligence Principles.
   • The FAQ explains that if a vendor contract does not allow insurance regulators to verify information from the vendor, the contract may be void for public policy reasons. In any case, the information used by insurers is subject to the regulatory authorities of participating countries.
3. Workstream #2 of the Big Data and Artificial Intelligence (H) Working Group, through the Draft Models and Data Regulatory Questions for Regulators, provides regulators with the right tools to ask insurers about their models and data We aim to At the group’s March 22 meeting, Commissioner Doug Omen recognized the need for changes based on comments received from stakeholders on the draft question. The draft question review includes the following subjects:
   • The scope of the question and the need for a limited, principles-based approach to foster innovation without over-regulation, including overlapping regulation.
   • Compliance costs for small businesses and vendors.
   • The impact of imposing liability on third-party vendors on insurers.
   • The need to protect confidential vendor information.
   • How to test data and models.
   • Based on the comments received, Mr. Omen said an amendment should be drafted by the end of May.
4. The Accelerated Underwriting (A) Working Group has issued a draft guidance document as another tool to assist regulators in reviewing accelerated underwriting programs used by life insurers. This tool provides sample questions and areas for regulators to review as they prepare the food they serve to insurers. The draft includes the NAIC’s Artificial Intelligence Principles, among others, in a framework of questions to facilitate regulatory assessments of whether accelerated underwriting programs are fair, transparent, and safe, as required by existing law. Various elements are included. The group explained that:

The use of accelerated underwriting affects both the availability and affordability of life insurance to consumers, so it is important to ensure that the use of accelerated underwriting is fair to consumers. Enabling insurers to take advantage of expedited underwriting in a transparent manner, as consumers need to understand what personal data insurers have access to and how that data is being used. It is important to Finally, insurers with access to sensitive consumer data have an obligation to protect that data in order to protect consumers from harm from unauthorized disclosure.

Colorado Department of Insurance

1. Prior to the February 7 stakeholder meeting, the Colorado Department of Insurance released a draft “Governance Rules for Algorithms and Predictive Models” and invited informal comments on the draft. Informal comments are now available on the department’s website. The draft regulation appears to be still under review as the department has taken no further action.
2. On April 6, 2023, the department held its first stakeholder meeting on private passenger auto insurance. This meeting was an appetizer for the main course to come. Specifically, the department reiterated the goals of the law and introduced the problems posed by insurers’ use of external data and AI in the context of private passenger motor insurance. The department opened the floor to questions that would make stakeholders bite the onion. Stakeholders provided feedback on the bill, but the main concerns focused on moving too quickly and the final dish being undercooked.

Other states

1. new york

In January 2019, the New York State Department of Financial Services issued a circular notifying insurers authorized to handle life insurance in New York State of their legal obligations regarding the use of external consumer data and information sources in underwriting life insurance. Issued the first book. The letter also requires insurers to (1) determine that their external tools and data sources do not collect or use prohibited standards and (2) ensure that their underwriting and rating guidelines are not unfairly discriminatory. It also seeks to establish that there is no The letter went on to say that insurers “cannot rely solely on vendor nondiscrimination claims or the idiosyncratic nature of third party processes as justification for their inability to independently determine compliance with antidiscrimination laws.” I am warning you. The burden is always on the insurance company. ”

2. Connecticut

On April 20, 2022, the Connecticut Department of Health issued a Notice on Using Big Data and Avoiding Discriminatory Practices. The notice was intended to remind insurers that they expect them to comply with anti-discrimination laws in their use of technology and big data. The department will provide insurers and third-party data vendors, model developers and secretariat access to the data used to build the models and algorithms contained in rates, forms and underwriting returns. discussed the authority to request The department emphasized the importance of data accuracy, context, completeness, consistency, timeliness, relevance, and other key elements of responsible and secure data governance.

3. California

On June 30, 2022, the California Department of Insurance released a bulletin titled “Allegations of Racial Prejudice and Unfair Discrimination in Marketing, Ratings, Underwriting, and Claims Practices by the Insurance Industry.” It imposes an obligation on insurance companies to “sell and issue insurance, charge premiums, investigate suspected fraud, and pay claims in a manner that treats all similarly situated persons in the same manner.” reminded me. “Conscious and unconscious prejudice and discrimination can be attributed to artificial intelligence and other forms of ‘big data’ (i.e., very large data sets that are analyzed to reveal patterns and trends),” the ministry said. It can, and often does, result from use.” He cautioned that the use of algorithms and models must have a good actuarial relationship with the risk of loss. The report goes on to say that “even where models and data suggest an actuarial relationship to loss risk, discrimination against protected class individuals will not occur unless a specific law expressly provides otherwise.” It is categorically and unconditionally prohibited.”