This AI Insurance policy covers claims filed by your customers for bias in your AI.
Most of the AI Insurance policies such as AI Performance Guarantee helps in gaining your customer's trust, hence, the right time to buy AI Insurance would be before you start your sales efforts or business development efforts or marketing efforts.
Yes, NLP use cases can be insured. An inbound sales automation company that uses NLP can buy AI Insurance because they handle millions of dollars of portfolio, hence, want to provide a financial backing if something goes wrong. A Gen AI company can buy AI Insurance to guarantee fairness and factual correctness of the generated text.
The payouts are decided during underwriting and the premiums vary accordingly. For example, a battery analytics company pays upto 2-3x of the price they have charged to their customers. For a payment fraud detection company, the payout is the exact amount of financial loss that their customers have suffered because of un-detected fraudulent transaction.
During underwriting, five different categories of metric are taken into account - data quality metrics, annotation metrics, training metrics, production setup and the process. The process starts with an introductory call where the business use case and required level of performance to guarantee are determined.
An AI need not be 100% performant to get an AI Insurance. The AI must be performant enough to not cause loss that might make the business use case unviable. For a AI payment fraud detection company, an accuracy of 20% might be good enough while for an AI cyber-security company maybe a performance close to 100% might be needed. These are very broad example, more context is needed to answer the acceptable threshold for performance.
For a Generative AI company, an example of performance metric may be factual correctness. The threshold maybe 100% and in this case if the Gen AI produces any factually incorrect text, the loss will be covered. A relevant metric for speech transcription can be Words Error Rate (WER). For a predictive analytics AI, the performance can be accuracy or recall of the predictions, whichever is most crucial in the context the AI is being used.
Generally, an AI company chooses a performance metric that is most relevant to their users/customers.
Opting for standalone AI insurance, instead of bundled Tech E&O, enhances sales as it becomes a key part of the sales pitch.
AI Insurance covers financial losses due to AI perils while Products-Completed Operation Insurance covers bodily injury and property damage due to use of product/completed work outside the business premises.
The underwriting process for AI Insurance is not standardised unlike Products-Completed Operation Insurance. The insurers require details around data controls, training and deployment that is unlike the underwriting process of Products-Completed Operation Insurance.
The Tech E&O provides liability coverage (third-party coverage) for professional services such as software consulting and might not cover the financial losses caused by the AI developed by your company. The E&O policies provides settlement when negligence can be proved and it might be difficult to prove negligence in case of AI model under-performance. An AI SaaS company might not be obliged to pay for a copyright infringement claim or fairness claim because of their end user agreement. Similarly, an AI consulting company might not be a held liable because of the terms of the contract.
The failure of an AI system to perform as required is usually not considered as a cause of business interruption, hence, Business Interruption Insurance will not cover business income losses due to AI perils.
The covered perils in Cyber Insurance usually are cyber-attacks and other data breaches related events. The AI perils such as model under-performance, model bias and model copyright issues are not covered perils in Cyber Insurance.
Content generated using generation AI such as text, music, images and other digital/physical assets might infringe upon a copyright protected asset. This scenario might not be covered by IP insurance.