Fide AI

Research Agenda

A research agenda for trustworthy AI.

Explore questions spanning model training, system design, evaluation, and human judgment. Each theme connects published evidence with open questions for further research.

Themes & domains

Questions connect. Context matters.

Research themes describe what we investigate: capabilities, alignment, evidence, human agency, governance, and work. Domains describe the settings in which those questions arise. Faith and religious life has its own research hub. Future directions and selected external work are presented separately.

Our mechanistic interpretability calls connect several themes: investigating model internals, improving behavior through training and system design, and testing whether the evidence helps people judge AI.

Explore mechanisms of trustworthy AI behavior →

Areas of inquiry

7 themes connect the work.

Each area names a durable set of questions. Together they cover technical behavior, evaluation, human consequences, institutional decisions, and the questions that recur across application settings. Published studies and open research calls turn them into work that can be tested, criticized, and extended.

01

Frontier Capabilities and Emerging Risks

What new risks appear as systems become more capable and autonomous?

We measure behavior across long-horizon, autonomous, and high-consequence tasks, including failure modes that existing safeguards may not address.

02

Agent Alignment and Runtime Assurance

Does an agent remain aligned while it plans, uses tools, and acts?

Our agenda connects model learning with system design: training data, objectives, feedback, and environments alongside permissions, operational traces, escalation, intervention, and recovery. We seek collaborators to test training methods.

Read the agent assurance research call →
03

Evaluation Science and Technical Assurance

Can the evidence support the decision people want to make?

We develop benchmarks, audits, red-team methods, reviewer protocols, and full-system evaluations that produce valid evidence about behavior and safeguards.

04

Grounding, Truthfulness, and Epistemic Integrity

Will the system use the right source and represent it faithfully?

We test retrieval, citation, provenance, reference selection, honest uncertainty, and the path from an authoritative source to a final answer.

05

Human Dignity, Moral Agency, and Formation

What judgment, habits, and relationships does AI use shape?

We study dependence, responsibility, attention, relational substitution, and formation over time, alongside tools that help people judge AI evidence and decide whether to rely on a system.

08

Stewardship, Governance, and Institutional Readiness

Which safeguards and authority structures should govern deployment?

We study how measured capabilities and risks should inform procurement, oversight, accountability, deployment decisions, and standards for responsible adoption.

09

Work, Vocation, and the Common Good

How does advanced AI change useful work and human contribution?

We study economic participation, professional responsibility, technological displacement, access to AI benefits, and the meaning people attach to work.

Continue into the research.

Read completed studies, inspect open questions, or contribute to a defined project.