The latest insights from Reinsurance Group of America (RGA) highlighted how DeepSeek’s open architecture could reshape AI adoption in insurance and risk management.
DeepSeek, a Chinese artificial intelligence company, introduced its multimodal large language models, DeepSeek-R1 and DeepSeek-R1-Zero, on Jan. 20, positioning itself as a competitor to OpenAI’s o1 model.
DeepSeek’s latest models follow the December release of DeepSeek V3, a language model comparable to OpenAI’s GPT-4. The R1 model, in contrast, focuses on reasoning, aligning with OpenAI’s o1. China has been increasing its presence in AI research, with models such as Alibaba’s Qwen series contributing to advancements in the sector.
Get the latest reinsurance news direct to your inbox twice a week. Sign up here
RGA said that intense domestic competition has fueled a price war, reducing the costs associated with hosting large language models and accelerating technological development.
China’s rapid AI progress is influenced by two major geopolitical factors. First, restrictions on Western AI models have driven domestic innovation. As of January, OpenAI’s GPT models and Meta’s Llama models remain either blocked or officially unavailable in China due to regulatory policies and provider decisions. This has led Chinese firms to focus on developing proprietary models.
Second, US export controls limit China’s access to high-performance Nvidia graphical processing units, which are essential for training AI models. Despite these constraints, DeepSeek reportedly trained its V3 model, comparable to GPT-4, using only 5% of the GPUs OpenAI utilized. RGA highlights that these efficiency improvements could lower the computational costs associated with AI model development and fine-tuning.
Meta’s chief AI scientist, Yann LeCun, recently stated that open-source AI models are surpassing closed ones. This distinction is critical for insurance companies, as transparency, interpretability, and cost-efficiency influence AI adoption. Closed models, such as OpenAI’s GPT, Google’s Gemini, Anthropic’s Claude, and AWS’s Titan, maintain proprietary training data and architecture.
Read More: RGA makes strategic investment in PACT Capital to expand alternative assets
These models are accessible only via APIs, limiting insurers’ ability to audit internal processes and raising concerns about compliance and data security. Open-source models, including Meta’s Llama, Mistral, and DeepSeek’s offerings, provide access to trained weights, enabling insurers to deploy and customize models within their own infrastructure.
RGA said that this flexibility improves compliance with regulatory frameworks governing data privacy and risk assessment in insurance.
Extracting insights from unstructured medical data is a priority for life insurers. Underwriting requires processing medical histories, physician notes, and diagnostic reports. While general-purpose large language models can assist, fine-tuning models tailored to insurance-specific needs may be more cost-effective than relying on API-based closed systems.
Open models allow insurers to train AI within secure environments while adhering to industry regulations such as HIPAA and GDPR. RGA points to DeepSeek’s ability to train V3 using a fraction of the typical computational resources as an indicator that fine-tuning and training costs may continue to decrease.
The emergence of open-source AI presents opportunities for life insurers to reduce costs, improve accuracy, and maintain control over sensitive data. DeepSeek’s advances suggest that AI innovation is shifting toward open ecosystems, providing insurers with greater flexibility in model deployment and customization.
As AI adoption in insurance grows, leveraging open-source models can help companies enhance operational efficiency while ensuring compliance with data privacy regulations. RGA said that insurers prepared to integrate open AI solutions will be positioned to optimize underwriting processes, manage risk more effectively, and drive business growth in a competitive market.
What are your thoughts on this story? Please feel free to share your comments below.
Get the latest reinsurance news direct to your inbox twice a week. Sign up here
