Research & innovation
Applied research, held to engineering standards.
Research themes
What we’re investigating.
Human-in-the-loop AI
Designing AI systems in which qualified people set the goals, review recommendations and remain accountable for outcomes.
Explainable, auditable recommendations
Making AI-assisted recommendations traceable: what informed them, who approved them and how they performed over time.
Retrieval-grounded generation
Techniques for grounding language models in curated, controlled knowledge to reduce unsupported output.
Evaluation and governance
Methods for testing AI behavior before release and governing models and prompts as they change.
Secure AI integration
Threat models and controls for AI-enabled applications, from prompt injection to data leakage.
Financial data intelligence
Applying analytics to consolidated financial data while preserving accuracy, privacy and access control.
Research principles
How we conduct and share research.
Claims follow evidence
We describe research by its status and method. Results are reported only once they are validated.
Privacy by design
Research systems minimize the data they collect and protect what they hold from the outset.
People stay accountable
In sensitive domains, AI supports qualified professionals — it does not replace their judgment.
Responsible disclosure
Project details are shared publicly only when they are ready and approved for release.
Initiatives are published when they are ready — and approved — for release.
Our research initiatives are in active development. We share project details, methods and findings publicly only once they reach an appropriate stage, with accurate status labels and without claiming results that have not been validated.
Start a conversation
Interested in research collaboration?
We welcome conversations with organizations and researchers working on trustworthy, human-centered AI and secure systems.