Lessons from the Frontlines of AI in Compliance

• Privacy, Security, and Governance Are Non-Negotiable

In compliance, privacy and security are foundational. Implementing localized data processing environments, as was done with AskFIN, ensures sensitive information is safeguarded and does not feed external models. Permission-based access controls add another layer of security, aligning with the unique demands of the compliance space.

• Transparency is Critical

Transparency around sourcing and function is not just appreciated—it’s required. A “black box” solution erodes trust, increases user confusion and questions, undermines adoption, and adds strain on client support teams. Providing context about how answers are derived and linking them to credible sources fosters accountability and confidence. Even after initial implementation, organizations must continuously iterate to maintain transparency, trust and confidence.

• Establish Guardrails for Responsible Use

Robust guidelines and guardrails are essential to manage the dynamic capabilities of gen AI. These measures help ensure responsible AI application and mitigate potential harm. For example, AskFIN has implemented guardrails to prevent questions outside the financial integrity domain while maintaining flexibility for users to introduce new terms within the model’s learning scope.

• Ongoing Testing and Quality Assurance

Continuous validation and iterative improvements are vital in the dynamic gen AI landscape. For example, with AskFIN, ongoing testing of its knowledge repository, architecture, and underlying models ensures its capabilities remain current and effective. This commitment to quality assurance builds resilience in rapidly changing environments.

Unique Implementation Strategies

To deliver trusted gen AI solutions like AskFIN, the following approaches were adopted:

• Collaborative, Multi-Disciplinary Development

We embraced a collaborative, multidisciplinary approach to implementing gen AI, bringing together the expertise of in-house subject matter experts (SMEs), technical and product teams, and valuable insights from our clients. This approach ensured a well-rounded perspective that addressed every aspect of the implementation process. Actively seeking out diverse viewpoints allowed us to refine functionality, tackle emerging challenges, and drive innovation to create a solution that truly resonates with users.

• Red Teaming for Innovation and Risk Mitigation

We engaged a second development team to “red team” our process, using a creative approach to identify opportunities for rapid improvement and uncover vulnerabilities to address. Much like a penetration test designed to challenge boundaries and think outside the box, this skilled external team applied their expertise to critically evaluate and expand our mental models, ensuring a more robust and innovative outcome.