How generative AI could help insurers with cybercrime

The rapid rise of generative AI has been a blessing and a curse for insurers offering cyber policies. So far, we've heard more about the latter. From deepfake images that have introduced entirely new categories of risk to supercharged password cracking and CAPTCHA-breaking technologies, the weaponization of generative AI has created some significant new challenges for insurers trying to build profitable cyber insurance practices. Fortunately, the technology is also making it easier to safeguard against these types of attacks and assess cyber risk more accurately.

In fact, the introduction of new AI-powered cyber risk analytics and simulation technologies may just be the key to finally making cyber insurance as profitable as it should be. Until now, that has not been the case. While cyber insurance loss ratios declined to 43% last year from 68% in 2021, the segment has continued to present challenges for insurers. 

Insuring a volatile risk class

Unpredictable patterns of behavior, a constantly-changing landscape of geographic hotspots and ever-evolving methods of attack have made the space incredibly volatile. Allianz Global Corporate & Specialty's Risk Barometer recently ranked cyber risks, such as IT outages, ransomware attacks and data breaches as the number one risk to businesses today.  Additionally, IBM's annual Cost of Data Breach report put the average cost of a corporate data breach at $4.35 million.

Accordingly, even though corporate interest in cyber insurance policies has been steadily rising, the actuarial science behind those policies has struggled to keep up, leaving the industry with inconsistent methods of benchmarking, tracking and reporting risks and—as a result—inconsistent performance. . 

AI is changing that paradigm by providing insurers with the tools they need to accurately model myriad cyberattack scenarios and develop more precise risk assessments. Based on our work with some of the world's leading P&C insurers offering cyber policies, we've found that the biggest challenge many encounter when it comes to reaping the benefits of this AI-driven approach is a lack of sufficient historical cyber data, which hampers their ability to make well-informed decisions. However, by pulling together the right mix of proprietary data and third-party sources, it is possible to build an integrated dataset that provides invaluable insights for predicting the likelihood and severity of future claims. Following are some of the specific areas where we are seeing the most rapid advances among firms that are adopting this approach:

• Generating synthetic data: Generative AI algorithms are being used to generate synthetic data that resembles real-world cyber threats, including malware samples, phishing emails, and network attack patterns. This synthetic data can then be used to train machine learning models, enhancing their ability to detect and classify new and evolving threats.

• Anomaly detection: Insurers are currently deploying AI to detect anomalies within network traffic, system logs, and user behavior by establishing baselines from normal patterns. By generating synthetic data that mimics legitimate network traffic or user behavior, any deviations from these patterns can be identified as potential indicators of a cyber threat.

• Simulating attacks: Simulated attacks, mimicking real cyber attacker behavior, are being reconstructed with AI models to help security teams proactively search for vulnerabilities within their systems, networks and applications. By analyzing the generated attack scenarios, organizations can identify holes in their current security and develop appropriate countermeasures.

• Threat intelligence sharing: Generative AI is also being used to anonymize and aggregate sensitive cyber threat data, allowing organizations to share information with trusted partners or security communities. By generating synthetic data that conceals the original sources, organizations can contribute to a collective knowledge base without compromising their own security.
• Malware detection and analysis: By analyzing a wide range of features such as disk access, APIs, bandwidth usage, processor power, and internet data transmission, generative AI is being used to identify and analyze various types of malware, including viruses, Trojan horses, worms, exploits, botnets and ransomware.

The list goes on and continues to grow every day. Ultimately, by deploying AI-powered analytics to continually assess the multitude of constantly changing risk factors businesses are facing every day, insurers are developing increasingly sophisticated approaches to managing—and accurately pricing—cyber risk. Other areas where AI can help transform cyber risk insurance operations include providing a consolidated score for an organization's overall cybersecurity posture, incorporating various parameters including technology stack, risk signals at an internet scale, topology, threat level, business priorities, regulatory obligations, and historical insights. 
It is still early days in the generative AI revolution, and for every step forward cyber risk professionals take toward securing their clients' networks, bad actors are chasing close on their heels with creative new ways to exploit weaknesses. This cat-and-mouse game will likely never end, but, with the right tools, insurers will be able to build a more predictable business at the center of this chaos.