AI, Risk, and the Public Sector: Navigating the Insurance Maze
A Coalition member shares their experience having to suddenly secure AI insurance for their chatbot, cautioning other agencies to check their own liability coverage for AI.
By Robert Hiett
Since November 2022, progress in artificial intelligence (AI) has felt like we are running a sprint and a marathon simultaneously. Each day, I learn about new technological approaches to solving problems, operating platforms to make staff more productive, and tools to help citizens navigate government programs and services. I, like others in the public sector, find myself busy planning and coordinating AI initiatives, and thinking about a future filled with amazing AI solutions that drive community and economic development, regional prosperity, and innovative workforce development solutions. Little did I know that “insurance” was going to jump out of the virtual bushes as an issue, and threaten to derail my entire AI operation. I’ll share a bit of our story with you, but feel free to skip towards the end if you just want to get to the insurance part of the article.
The Upper Coastal Plain Council of Governments (UCPCOG), with 46 member governments, is located in Eastern North Carolina about an hour away from Raleigh, and our region has a mixture of rural, suburban, and small metropolitan areas. The region is experiencing growth in advanced manufacturing, healthcare, transportation and logistics, technology, and agriculture. However, the UCPCOG also has one of the fastest-growing senior citizen populations in the state, and demand for resources and services is expected to grow over the next 20 years.
I am a believer in technology. Over my career, I have witnessed the evolution of technology, from typewriters and fax machines to personal computers becoming household staples, the rapid expansion of the internet, and now the emergence of artificial intelligence as a transformative force by the end of 2022. I have a history of working with technology to do innovative things, and my last project had been working with game-based simulations, career and technical education (CTE), and local area high schools to develop new workforce career pathways for in-demand occupations in the UCPCOG region.
I looked at AI as another opportunity to leverage technology to address an issue in the region. Not only is our senior citizen population growing, but more of our working-age population is beginning to find themselves carrying out the role of “caregiver” for an elderly parent or loved one. If you haven’t experienced this before, it can be a stressful time for the senior citizen who is navigating the challenges of growing older, a companion or spouse who is trying to support his or her loved one, or a caregiver who may be trying to juggle kids, work, job training, and trying to find governmental programs or services. Enter the AI solution, AskAshleigh, a truly generative AI-powered chatbot that can do more than just read from a script!
While I was developing that partnership and working to design the solution there, it became obvious that a “digital divide” related to AI was happening in the public sector. Any public sector leader who has been part of a broadband planning or digital literacy planning group can immediately draw the connection to the words “digital divide.” In the AI landscape, it means that most, not all, of the public sector enterprise-grade solutions, with all of their amazing tools, are happening in the highly populated areas of the United States. There just isn’t much out there for smaller local governments that are affordable, scalable, and secure. However, I was able to strike up a partnership with an AI provider to solve that problem and launch a regional initiative that has brought enterprise-level AI to the UCPCOG and its member governments.
All of this sounds great. We are “cooking with gas” as my grandfather used to say (he grew up in the Great Depression era), and I had my focus on the AI launch dates to roll out the initiatives. The UCPCOG was in the process of renewing its insurance policies because of some new regulations in North Carolina that required a specific type of coverage, and we were in the middle of going through the application and quote process when a few pieces of news got my attention. The news items were related to someone filing a lawsuit against a provider of a chatbot service, and some other news about students being sanctioned by education institutions for “allegedly” using AI to write content.
There are different types of chatbots out there. I have multiple AI certifications, so I am familiar with how they are designed and built, and there are good ones and bad ones. Regarding the lawsuit, the company in question developed a chatbot that seemingly provided incorrect advice. In the other case, experts widely acknowledge that AI detection lacks accuracy, making it unreliable for definitive assessments. Even OpenAI has refrained from implementing a tracking system due to its ineffectiveness. If an AI detector is introduced, it is likely that a workaround will emerge within weeks. AI writing and detection could matter if you or someone in your agency used AI to produce something that was inaccurate, misleading, or harmful, and perhaps you could find yourself in court over it.
Given what I had seen in the news, I reached out to the insurance companies we were working with on quote development for renewals, one being our current carrier, and I asked if I needed to increase my general liability and/or cyber insurance coverage. I brought up what I was hearing in the news because my question seemed to catch the brokers/carriers off guard, and I was told that AI operations were not covered under any policies they carried. I was pretty shocked to hear this answer, and I asked them to expand their search to see if they could help us find partners who could write AI policies.
One insurance group simply told me that they don’t provide AI insurance and weren’t planning to offer it anytime soon. Other brokers simply took too long to get back to me. When they did show up, they had too much paperwork that was going to take too long to fill out. Quite frankly, it just didn’t seem like any of them had a real plan for how to find the right type of AI insurance policy. If insurance brokers and underwriters don’t know what to do, then what does that mean for the rest of us?
During my hunt for AI insurance, I came to understand that many insurance brokers and underwriters don’t truly seem to grasp all the aspects of what AI systems are, how to define them, and how to assess risk. So the result is that, at the moment, there do not appear to be many brokers and underwriters out there willing to insure it. I even heard one phrase “assume it is covered unless we say it isn’t”, and I’ll remind you all that our policy was canceled because I asked a question about AI.
I immediately started a search for AI insurance providers. Neither Google Search nor Perplexity.ai was helpful in the process because they would confuse “AI insurance” with “cyber insurance”, and I learned that those two things don’t always mean the same thing.
When you are on a 30-day cancellation notice, I’m talking to the insurance people reading this article, you don’t have time to deal with a complicated, script-based chatbot that cannot provide real assistance or refer you to a human, and you don’t have time to waste filling out online questionnaires to send into the internet void while hoping someone will call you back shortly. You need to talk to a real human, literally, right freaking now. AI and technology are great unless they prevent you from reaching a live human when you need it.
During my search, I ran across the Allen Thomas Group. Joe Race (email: jrace@atgins.com), called me back promptly to see how he could help the UCPCOG. They were helpful and understood our needs, and we worked with them to obtain AI insurance coverage for our operations.
The typical cyber policies tend to protect you from data breaches, malware and hijacking attacks, business interruption events, and things of that nature. When shopping for cyber liability insurance make sure the cyber policy covers more than just standard email breaches and network attacks. It should include things like deepfakes and system failure, and extend to cover things related to contracted professional services and solutions.
The AI operations side gets tricky if you do it in-house. It appears to be more clean-cut and easy to follow if you are using professional services provided by a technology consultant or a technology company. You are going to want to ensure you are covered against AI bias, things related to algorithms, data sources used, and quality and continuity of services.
If you are a public sector leader and you have deployed or are planning to deploy any AI, internal, external, or both, then I encourage you to check with your insurance broker. Be sure to check your policies and endorsements package, and look for specific language around technology, professional services, software and systems, cyber liability, etc. If you are unsure what your endorsements truly say, then get written clarification from your brokers and underwriters as quickly as possible.
In conclusion, I clearly want to state that the UCPCOG did not lapse any insurance coverage. However, I find it unsettling that technology, such as AI, can roll out across the country, and insurance providers may not be giving public sector organizations advice and guidance on whether or not they have coverage.
About the author
Robert Hiett is the Executive Director of the Upper Coastal Plain Council of Governments in North Carolina.
Note: The opinions expressed in these articles are solely those of the author(s) and do not necessarily reflect the views, positions, or policies of the GovAI Coalition or the authors’ affiliated professional organization(s).
—
This is the blog for the GovAI Coalition, a coalition of local, state, and federal agencies united in their mission to promote responsible and purposeful AI in the public sector.
