InsurTech firm Akur8 is taking a controlled approach to AI adoption in insurance: if using AI for actuarial work is possible and even necessary today, it requires stronger safeguards than the “move fast” mentality often associated with artificial intelligence.
In a recent Akur8 analysis, Chief Operating Officer Felix D’Alançon outlined the challenges facing insurers as they introduce AI into actuarial workflows. He argued that errors in actuarial models can remain hidden for years, while AI still struggles to intuitively replicate the contextual judgement built through years of actuarial experience.
There is also a regulatory challenge. AI-generated outputs used in rate filings, reserve calculations or capital models make transparency, auditability and explainability essential.
Akur8’s answer is what it calls “Constraint Engineering”, a framework designed to keep AI operating within clearly defined boundaries. Humans establish the architecture, constraints and quality standards, while AI generates outputs within those parameters.
Rather than requiring users to inspect every AI-generated result, the framework focuses on controlling how systems operate. Automated checks can then test outputs against predefined requirements before they reach human reviewers. Akur8 said the approach delivered productivity gains of up to 50% when used by its own engineering teams.
The company has applied the same principles to its actuarial AI strategy through two layers of control. The first combines governed data, business logic and institutional knowledge with security and compliance infrastructure. The second provides a deterministic execution layer, using actuarial models, KPIs and logical checks to identify incorrect outputs before they reach users.
This approach underpins Akur8 Agents, which launched in the second quarter of 2026 and is integrated into the Akur8 platform.
The company said the agents are curated and validated by actuarial experts, with prompts and logic reflecting real-world actuarial practices and edge cases. Their workflows are also designed to be transparent, with activity logged and timestamped to provide an audit trail.
Akur8 said its agents run on its own cloud infrastructure and proprietary statistical engines, while prompts and data are not reused to train models.
The firm is now running workshops with actuarial teams to explore responsible AI adoption, highlighting a wider challenge for insurers: how to capture the productivity benefits of AI without compromising the controls required in high-stakes, regulated environments.
Copyright © 2026 FinTech Global
