All innovation has the underlying motivation of making human life easier, comfortable, and more efficient. In the process, the associated risks also transform. Risk, like energy, never disappears. It only changes form and magnitude.

Artificial intelligence (AI) and associated innovations are no different. While the usage of AI reduces the operational risks of errors (especially with repetitive tasks), fraud risks by identifying anomalous patterns, and accident risks for autonomous cars, it brings additional risks along with it. These include:

1. Legal Liabilities: AI systems can lead to legal accountability for harm caused by their actions or decisions, such as violating laws or causing damage. Example: An autonomous vehicle’s AI misinterprets a stop sign, causing an accident, leading to lawsuits against the manufacturer for negligence or product liability.
2. Algorithmic Error: Mistakes in AI algorithms can produce incorrect outputs, leading to flawed decisions or actions. Example: A credit scoring AI miscalculates risk due to a coding error, denying loans to qualified applicants, resulting in lost business and reputational damage.
3. Performance Failures: AI systems may underperform or fail to meet expected standards, affecting reliability and outcomes. Example: A medical diagnosis AI fails to detect cancer in scans due to poor training, leading to delayed treatment and patient harm.
4. Business Interruption: AI system failures or downtime can disrupt business operations, causing delays or lost revenue. Example: An AI-driven inventory management system crashes, halting warehouse operations and delaying shipments for an e-commerce company.
5. Data Breaches: AI systems handling sensitive data are vulnerable to hacking or unauthorized access, compromising personal or corporate information. Example: A healthcare AI platform is hacked, exposing patient records, leading to privacy violations and regulatory penalties.
6. IP Infringement: AI may unintentionally use or reproduce copyrighted material or proprietary algorithms, violating intellectual property rights. Example: An AI-generated music tool creates a song that closely resembles a copyrighted track, triggering a lawsuit from the original artist.
7. Financial Losses: AI failures, errors, or misuse can lead to direct or indirect monetary losses for businesses or individuals. Example: An AI trading algorithm misinterprets market signals, executing erroneous trades that cost a hedge fund millions in losses.
8. Biased Decisions: AI systems trained on biased data can produce unfair or discriminatory outcomes, perpetuating inequalities. Example: A hiring AI favors male candidates over female ones due to biased training data, leading to gender discrimination lawsuits.
9. Cyber Risks: AI systems can be targeted by cyberattacks, such as adversarial attacks that manipulate inputs to cause errors or system compromise. Example: Hackers feed altered images to a facial recognition AI, bypassing security protocols and gaining unauthorized access to a facility.
10. Regulatory Non-Compliance: AI systems may fail to adhere to industry regulations or legal standards, resulting in fines or sanctions. Example: An AI-powered chatbot collects user data without proper consent, violating GDPR, leading to hefty fines for the company.

This is clearly not an exhaustive list and the presence and impact of each of them would vary based on the use case of AI. Equally, all of these risks exist even without AI and in some form or shape. Most of these are insurable as well.

So, what are the AI risks that are being covered as of now?

The development of specialized AI insurance products represents a targeted response to the unique risks posed by artificial intelligence (AI), such as algorithmic bias, performance failures, and intellectual property (IP) infringement. Providers like Munich Re, Vouch, Armilla Assurance, and Superscript have pioneered these offerings to address liabilities that traditional insurance policies, such as cyber insurance, professional indemnity, or third-party liabilities, often fail to cover comprehensively. 

These products are evolving rapidly to align with emerging regulatory frameworks, such as the EU AI Act, and to meet the needs of businesses deploying AI across industries.

How is the risk associated with AI loss assessed?

Risk due to AI is very specific to the use case. What may be a high risk in the healthcare environment could be a low risk in a financial advisory use case. Insurers use actuarial models, machine learning, and predictive analytics to estimate the probability and severity of AI-related losses. This includes:

• Quantitative Modeling: Estimating financial impacts of risks like data breaches (e.g., Equifax’s $425M settlement) or business interruptions (e.g., downtime costs for an e-commerce AI chatbot).

• Qualitative Analysis: Assessing non-quantifiable risks like reputational damage from biased AI decisions or regulatory fines for non-compliance with laws like GDPR or the EU AI Act.

• Specialized Considerations: AI-specific risks like adversarial attacks (manipulating AI inputs) or deteriorating model performance are modeled using emerging cybersecurity frameworks.

Pricing AI risk?

The premium methodology is more aligned with calculating risk for any other liability risk. For example, third-party risk in a motor insurance, i.e,. Risk of loss of third-party (human or property) resulting from an accident. 

Like any other risk pricing, AI risk is a combination of both science and art. In case of AI, the degree of art (which is a loose term for actuarial judgement) is significantly higher as compared to other traditional insurance products. This is calculated based on the assessed risk exposure, incorporating:

• Base Risk Premium: Reflects the expected loss from identified risks, adjusted for the AI’s use case (e.g., high-risk medical diagnostics vs. low-risk spam filters). Expected loss in turn is calculated based on the frequency and severity of each risk.

• Uncertainty Loading: Accounts for the unpredictable nature of AI risks, such as rapid technological changes or emerging regulations.

• Mitigation Discounts: Lower premiums may be offered if the insured implements robust AI governance, such as regular audits, explainable AI practices, or human oversight.

• Market Factors: Competitive pressures and the limited historical data on AI losses influence pricing, often leading to conservative estimates.

Challenges in assessing AI Risk?

Limited Historical Data: Unlike traditional risks (e.g., auto accidents), AI risks lack extensive claims data, forcing reliance on simulations and expert judgment. To that end, it is about pricing a cyber risk.

Evolving Technology: Rapid advancements (e.g., generative AI like ChatGPT) outpace underwriting models, increasing uncertainty.

Regulatory Uncertainty: Varying global regulations (e.g., EU AI Act vs. China’s AI policies) complicate compliance risk assessment.

Ethical Concerns: Bias and fairness issues are hard to quantify, yet critical, as seen in lawsuits like A.F. v. Character Technologies, alleging AI-driven harm to youth.

Systemic Risks: Potential catastrophic risks from AGI or AI-driven misinformation require insurers to consider low-probability, high-impact scenarios.

Conclusion

AI is here to stay and evolve. So are the risks associated with AI. There are early examples of reinsurers covering the AI-associated risks, which are largely confined to algorithmic bias, performance failures, and intellectual property (IP) infringement, but this is likely to get more comprehensive. Underwriting and Pricing AI risks continues to remain a challenge mainly because of the evolving technology, lack of historical data, varying and evolving global regulations.