01 · Our purpose
Building trustworthy AI
Research that shapes how models learn and systems behave—from training data, objectives, and feedback to system design and human oversight.
Independent AI research lab
How do we know when AI deserves our trust?
Measurement science for trustworthy AI.
We research how to build more trustworthy AI and help the public discern whether to trust it.
02 · Our purpose
Equipping public discernment
Tools and evidence that help people and institutions understand AI’s capabilities and limits, exercise their own judgment, and decide when to rely on it.
Selected research
View all research →Which changes improve AI behavior, and what evidence shows that the improvement is sufficient for a particular use? Explore what our studies found and the questions they leave open.
Study · Faith & Religious Life Knowing When to Defer
Access to a source did not ensure its use.
In a controlled study of Scripture quotation, we tested whether AI would consult an authoritative source when users asked it to rely on memory. We varied whether source use was optional or required.
User pressure reduced source consultation under optional instructions. A higher-priority source requirement preserved much more of it, but did not guarantee consultation.
Read Knowing When to Defer →Source consultation by condition
- Optional source · ordinary request
- 95.1%
- Optional source · user requests memory only
- 30.6%
- Required source · user requests memory only
- 84.7%
4,800 controlled observations · six model families · 20 passages. Findings apply to these study conditions. Transfer to other domains requires further testing.
01 / 04
Our approach
Understand behavior. Investigate what drives it.
Understanding AI means asking both what a model does and why. Our agenda connects behavioral experiments with mechanistic interpretability: studying the internal computations that shape a model’s responses and decisions.
We seek collaborators to investigate how models use evidence, weigh moral reasons, and respond to correction, and how that understanding can improve the systems we build.
- 01Evidence
- Do the records support the claims, and were the tests sufficient for the decision?
- 02Authority
- Is the system’s role legitimate, and does it respect the limits of that role?
- 03Human judgment
- Can people recognize mistakes, challenge conclusions and verify that their interventions work?
Independent evaluation · October 15, 2026
Independent evaluation is moving into practice.
Anthropic reports that METR now has employee-like access. Fide’s approach connects model behavior, the effectiveness of safeguards, and the people who rely on them.
What changed, with sources →Work with Fide
Three ways to put the research to work.
Different teams need different kinds of help: research assets to improve AI, independent evidence about a system, or specialist safety capacity. Choose the engagement that fits your work.
01 · Frontier labs and advanced AI teams
Research assets for AI developers
Need to teach and measure judgment in difficult domains?
Commission research on source fidelity, contextual judgment and human authority, with benchmarks, expert-reviewed data or interactive learning environments as agreed outputs.
You receive: A defined research question, protocol and agreed asset or study deliverables, with validation, limitations and usage rights specified.
Fide’s role: Research partner developing assets you use to improve AI
In development: New training datasets and reinforcement learning environments.
02 · Anyone who must rely on or answer for an AI system
Independent evaluations
Need an outside assessment of an AI system?
Evaluate a model, product, system or agent harness from outside the team that built it. Compare versions, investigate behavior, or examine whether an evaluation and its graders measure what matters, for a decision your leadership, customers or partners need to make.
You receive: Documented findings, supporting cases, methods and limitations, recommendations and a decision briefing.
Fide’s role: Outside evaluator. The findings are ours.
03 · Teams building AI without an in-house safety function
Applied AI safety
Need hands to find and fix problems?
Work alongside Fide on the system you are building: red-team it, harden it before release and build the tests your team will keep running. This is improvement work, not an outside opinion.
You receive: Failure cases and fix priorities worked through with your engineers, retesting of fixes, and test materials or a handoff where included in scope.
Fide’s role: Part of your team. The goal is a better system.
For the public
Public tools for understanding AI.
Explore Fide’s public tools, interactive research and plain-language Insights. Follow AI claims back to their sources, understand what remains uncertain, and decide what to verify or keep in human hands.
Explore all public tools
Open the Agentic Cyber Explorer
PreviewFollow how AI agents attack, defend and fail. Examine the findings and the evidence behind them.
Research domains
Trust takes different forms in different settings.
Explore our published work in faith and religious life, our cybersecurity investigations and developing autonomous-defense research, and questions taking shape in enterprise and other domains.
Fide AI · Published research
Faith & Religious Life
Religious sources, spiritual authority, moral judgment, and the responsibilities of people and institutions.
Our first domain of published empirical work, with studies in Christian contexts.
Explore Faith & Religious Life →Fide AI · Published analysis
Cybersecurity
Research toward dependable autonomous cyber defense: when to act, how to verify recovery, and how to preserve human control.
Explore investigations, current studies and longer-term questions about evidence, oversight and autonomous defense.
Explore Cybersecurity →Fide AI · Exploratory direction
Enterprise
Trustworthy AI for knowledge work: reliable outputs, accountable actions, and effective human review.
An exploratory direction, starting with how to evaluate AI-generated decision briefs against their sources.
Explore Enterprise →Research independence
Methods you can inspect. Judgments you can question.
Every research program reflects judgments about what matters and how evidence should be interpreted. We make those choices visible so readers can examine our methods and challenge our conclusions.
State the lens
Disclose the assumptions, prior commitments, and worldviews that materially shape each project.
Show the evidence
Make methods, findings and limitations available for scrutiny, while protecting confidential information and explaining material restrictions on disclosure.
Separate evidence from judgment
Distinguish measured findings from interpretation, recommendation, and moral judgment.
Fide retains control of its conclusions. Payment cannot determine favorable findings. We disclose material limits on access and what those limits mean for the conclusions we can draw.
Research participation
Help investigate questions that matter.
Universities, institutions, researchers and sponsors can help shape and pursue a substantive research agenda. Scope, rights, credit and funding are agreed for each collaboration.
Collaborate on research
Bring a research question, institutional setting, systems or expertise that could support a shared study.
Explore collaborationContribute expertise
Help develop and review methods with research, engineering, domain practice or community experience.
Join the workSupport a study
Fund a defined research question, agreed outputs and the people and infrastructure needed to investigate it.
Explore research support
Newsletter
Follow what we find—and what we test next.
Get new findings, the evidence behind them, and the questions guiding our next studies.