Research and technical engagements
Research assets. Independent evidence. Applied safety.
Work with an independent AI research lab on consequential tasks where sources, context, human judgment and institutional responsibilities matter. Commission technical work, develop your institution’s AI strategy or define a continuing engagement.
Where the work stands
Published today
Studies, benchmark materials and public tools, each with stated methods and limitations.
Available to scope
Commissioned research, evaluations, applied safety, institutional strategy and capability work. Founder-led, with expertise, qualified review, deliverables and fee agreed for each scope.
In development
New training datasets, reinforcement learning environments, packaged managed services and longer-term institutional AI infrastructure.
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.
Fide’s role: Research partner developing assets you use to improve AI
What you can commission
- Benchmarks and evaluation tasks
- A defined capability, task set, scoring protocol and validation plan. Hold-out and disclosure arrangements are agreed before development.
- Expert-reviewed data
- Cases, reference judgments or feedback with reviewer methods, disagreement, provenance and permitted uses documented.
- Reinforcement learning environments
- Interactive tasks and feedback for a specified capability, with reward validation and held-out assessment. This is a development direction to commission and validate, not an off-the-shelf environment.
What you receive
A defined research question, protocol and agreed asset or study deliverables, with validation, limitations and usage rights specified.
What to bring
The capability you want to improve, relevant domain or workflow, candidate models, available data and compute, intended uses and desired outputs.
Discuss commissioned research →Published studies demonstrate methods under their stated conditions. Data licensing, training rights, publication, exclusivity and any new asset development require an explicit agreement.
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.
Fide’s role: Outside evaluator. The findings are ours.
What you can commission
- Model, product, system or harness assessment
- A bounded behavioral question, realistic and adversarial cases, reviewed findings and the implications for your intended use.
- Comparison studies
- Evidence about changes to models, prompts, configurations or mitigations under a consistent protocol, including regressions and tradeoffs.
- Evaluation and grader validity review
- Examine rubrics, reference judgments, grader calibration, coverage or the evidence behind a benchmark result. Reconstructing a score does not by itself validate its meaning.
What you receive
Documented findings, supporting cases, methods and limitations, recommendations and a decision briefing.
What to bring
The system and version, decision you need to make, relevant criteria, available access and evidence, intended recipients and timing. Begin with non-sensitive context.
Discuss an evaluation →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.
Fide’s role: Part of your team. The goal is a better system.
What you can commission
- Focused red-teaming
- An agreed threat model, adversarial scenarios, reviewed failure cases, coverage limits and priorities for follow-up.
- Pre-release hardening
- Test, fix and retest with your engineers until the agreed risks are addressed or clearly documented for a release decision.
- Evaluation engineering and handoff
- Build a working evaluation harness, reviewed cases, grading criteria and regression checks your team can maintain. Scope includes documentation and team handoff, with grader validation and usage rights specified for the work.
What 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.
What to bring
The product or workflow, the safety question, your existing testing, the capability you need help with, available access and the people who will own the work afterward.
Discuss a safety engagement →Implementation help and independent assessment have different roles. Testing work Fide materially helps design or build is collaborative validation; an independent opinion on it needs another evaluator.
Customer-specific deliverables have agreed ownership and licensing. Fide retains its background software, general methods and reusable tools under their existing licenses.
Compare two paths
Outside assessment or hands-on improvement?
Both can inform a release, purchase or deployment decision. Choose based on Fide’s role: an outside evaluator with control of its methods and conclusions, or a specialist helping your team improve the system.
Independent evaluation
- Choose it when
- You need an outside assessment whose methods and conclusions Fide controls within the agreed scope.
- Fide’s role
- Outside evaluator. We select methods within the agreed scope and retain our conclusions.
- You receive
- Findings, supporting cases, limitations and a recommendation for the named use.
- Independence
- Required. We disclose any prior involvement with the system.
Applied AI safety
- Choose it when
- You need hands-on help finding and fixing problems, developing tests or improving the system.
- Fide’s role
- Part of your team. We work through failures and fixes with your engineers.
- You receive
- Fix priorities, retesting, and test materials your team can maintain and rerun under agreed rights.
- Independence
- Not independent. Where Fide materially helps design or build a system, an independent opinion on it needs another evaluator.
For companies, universities and institutions · Available to scope
Develop your AI strategy and capability.
What should AI do in your institution, and what should remain a human responsibility? Translate your purpose into practical choices about systems, evidence, data and authority.
Discuss your institution’s next decision →A useful first engagement
Begin with a defined decision or audience. A working session can support the work; the deliverable is an agreed plan or demonstrated capability.
- Strategy and roadmap
- Current uses, prioritized opportunities, human responsibilities, data requirements and a build/buy/partner comparison, with costed phases and local owners.
- Practical capability
- Hands-on work with realistic tasks, source and error checking, an assessment of what participants can do, and a local follow-up plan.
- A resourced next step
- A written recommendation and scope for a study, evaluation or bounded pilot where useful. The initial work should support a decision even if no further engagement follows.
Agree the audience, preparation, outputs, qualified expertise and fee before work begins. Technical implementation and infrastructure require separate scopes.
Continuing engagements · By agreement
Build on the work as your needs evolve.
Where there is a recurring need, agree a continuing research, evaluation or advisory relationship with a defined cadence and funded capacity.
Discuss a continuing engagement →Possible outputs include scheduled model comparisons, reassessment after consequential changes, maintained evaluation cases, technical decision reviews or milestones in a longer research program.
Specify the systems or questions, included work, review cadence, reporting, fee and renewal decision. Expansion and operational coverage are agreed separately.
Fide’s role stays explicit: outside evaluation retains independent conclusions; engineering support helps build or improve the system. An independent opinion on work Fide materially helps design or build needs another evaluator.
One scientific foundation
Understand judgment. Develop methods. Test their limits.
A domain research program can produce tasks, reference judgments and learning environments. Evaluations investigate what those methods establish about actual systems; applied safety work helps teams act on the findings.
We specify the sources, criteria, uncertainty and limits of authority relevant to each setting. Expert disagreement is evidence to investigate, and automated grades need validation against the question they are meant to answer.
Training material and held-out evaluation have distinct purposes. Private customer records carry no automatic research or training rights. An independent opinion on Fide’s own instruments and implementations needs another evaluator.
We agree scope, qualified review, reporting rights, timing and fee before work begins. A report supports a stated decision within its evidence limits; it does not confer certification.
Read our independence commitments →Longer-term direction · Sovereign AI
AI capability under institutional authority.
Research, evaluation and safety work can help institutions define their standards, understand system behavior and develop the ability to govern AI as it changes.
Our longer-term vision is to build and operate AI systems on infrastructure institutions control: their sources and data, chosen models, bounded workflows, human approvals and portable evidence.
In development: an integrated institutional platform and managed sovereign infrastructure. These are future research and engineering directions, not a currently available deployment service.
Explore a research partnership →Want to pursue a research question together?
Commission a study, co-develop a research question or support a shared program. Agreed outputs can include original studies, public resources and educational materials as well as technical assets.