Following the launch of ChatGPT in November 2022, the past three years have seen rapid, dramatic changes in the realm of GenAI, leading to advanced capabilities and increased awareness surrounding accuracy and accountability.
As 2025 comes to a close and I share my final generative artificial intelligence (GenAI) update of the year, it strikes me that it has been three years since ChatGPT launched on November 30, 2022. In this update, I’ll review the key developments over that time and offer a look at what could lie ahead in 2026.
Starting in early 2023, GenAI advanced quickly, with new models emerging each year, improving what these systems could do and how widely they could be used. OpenAI’s GPT-4 and GPT-5 led the way in language and reasoning. Quickly, Google’s Gemini, Anthropic’s Claude, and Meta’s LLaMA series offered strong alternatives to ChatGPT. DeepSeek took the world by storm, demonstrating that smaller companies could also create efficient, high-quality models.
For the insurance industry, this meant new opportunities in automation, document analysis, and customer service. However, it also raised important questions about accuracy, explainability, and accountability in AI.
Regulators worldwide began to take action. The European Union finalized the AI Act, which established specific rules for high-risk systems, including many used in insurance. In the US, the federal government used executive powers to guide AI oversight, requiring transparency and safety testing for advanced models. China set strict content and labeling rules for GenAI tools. These steps made it clear that AI was no longer something companies could work on alone. For insurers, this meant ensuring their AI strategies comply with new legal requirements, particularly for underwriting, pricing, and claims.
Several key themes emerged:
By the end of 2025, AI had become both a powerful tool and a shared responsibility.
We have already seen widespread adoption of GenAI in various computer applications. Likely, many of today’s software programs already support GenAI chatbots for automated use. This allows common tasks for underwriters, actuaries, and others to benefit from AI assistants.
Search is one of the most common first stops in the GenAI journey for insurance professionals. Just describe what you are looking for, and agents like Microsoft Copilot can scan meetings, emails, and instant messages that you shared with others in your company. Most company intranet sites now have GenAI search, which allows rapid lookups of company information.
Agents can empower multiple applications to work together. This allows individual users to create custom pipelines without the need to initiate a lengthy software development project. The agents can often generate code to perform more complex tasks. Agents will likely become even more powerful in 2026, allowing them to adapt to specific tasks.
Gazing further into the future, many are beginning to look for what the next advancement beyond large language models (LLMs) might be. Many AI luminaries wonder how far we can take LLM reasoning in terms of guarding against hallucination without being so cautious that the model overlooks a potentially creative answer.
AI pioneers Yann LeCun and Geoffrey Hinton shared the 2018 Turing Prize, the computer science equivalent of the Nobel Prize. Today, they have differing views on how LLMs might take us to the next major step in AI capabilities – artificial general intelligence (AGI), which enables AI to adapt to any problem. Hinton believes that LLMs might already be on the path to AGI, while LeCun believes they are a dead end. If Hinton is right, we may already be close to the next major AI breakthrough. If LeCun is correct, more research will be needed to discover a model that can deliver AGI. Either way, the next year will see AI move closer to AGI as models become more general and take on more abstract tasks.
Regardless, 2026 will likely be an exciting year for AI.
Have a question or comment for Jeff? Join the conversation.
