Research you
can audit,
for a complex world.
Oryx is an independent research initiative building governance-first methods for agentic AI in strategic research — where every material claim stays traceable to its evidence, and every consequential judgement stays with a human. It works with research institutions, funders and public agencies that need evidence they can defend.
Working paper v3.0 on Zenodo · Companion citation-audit study · Listed in the UN DESA Civil Society database
Move beyond plausible answers to research whose claims can be traced, checked, and defended.
Oryx AI turns the Centre’s framework into a service for research teams that cannot afford a confident-sounding error. One pilot report, pay only if it passes — the full offer and process live on the service page.
From question to consequence.
The Centre studies how AI systems are reshaping research, policy and statecraft — and builds methods for using them accountably, with a particular focus on the perspectives and priorities of the Global South.
Governance & evaluation of agentic AI
How agentic systems should be governed, and how their research outputs can be measured — sourced-claim rate, citation precision and recall, fabrication rate, calibration.
AI, geopolitics & technology competition
Strategic research on how advanced AI is redistributing capability and influence between states, institutions and regions.
Emerging tech & strategic foresight
Foresight and policy research for the Global South on the technologies — and the rules — that will shape the next decade.
Built to think in systems.
Our founding framework makes AI-assisted research verifiable rather than merely plausible: a six-layer process in which evidence is acquired, claims are traced, and humans keep the decisions.
Question framing
The research question is scoped and made answerable before any system runs.
Evidence acquisition
Sources are gathered deliberately, with provenance recorded from the start.
Claim-level traceability
Every material claim is tied to the specific evidence that supports it.
Adversarial review
Outputs are challenged — counter-evidence and weak sourcing are hunted down.
Human decision gates
Consequential judgements pause for an accountable human to decide.
Monitoring against drift
Systems are re-checked over time as models, sources and contexts change.
The founding paper.
The paper proposes a six-layer framework for making AI-assisted research verifiable rather than merely plausible, and defines measurable dimensions for agentic research systems — with an evaluation protocol for testing them. It is a conceptual framework with protocol targets, published openly for other institutions to test, reuse and challenge.
Practical checks from our research.
Short guides from the Centre's research on checking AI-generated claims and citations.
How to check an AI vendor's claims before you sign
A six-step check you can run on a vendor's claims — using the vendor's own materials — before you sign.
How to check AI-generated citations in your own paper
A four-verdict method for checking AI-generated citations, in about ten minutes per batch.
Pay only if it passes.
One auditable research report, on a question you choose. The pass criteria are agreed in writing before any work starts.
Send the question
You send the research question and the decision it informs.
Agree the criteria
We write the pass criteria with you, in writing, before any work.
Run and review
The six-layer process runs, and a human review gate checks the report before delivery.
Check against the criteria
You check the report against the criteria. Pass means an invoice is issued; fail means nothing is charged.
One revision round
One revision round is included in the pilot.
Everything on this site, in one glance.
Start with the service, read the research, or run a practical check yourself. Every page below is one click away.
Oryx is an early-stage research initiative founded on 2 October 2026 in Noida, India, by Amartya Kumar and Riya Nalwa. Pilot engagements are contracted through Riya Nalwa’s Udyam-registered proprietorship (UDYAM-UP-56-0189301). Nothing on this site claims a client, a funder or a result that does not exist.
Oryx AI
The applied service: one auditable research report, pay only if it passes criteria agreed upfront, with a guided demo and an enquiry form.
Sample report
A condensed sample from Oryx AI’s own launch test — labelled DRAFT, with its verification numbers and gaps in plain sight. Not a client delivery.
Working paper v3.0
The founding framework: from answers to auditable intelligence. 23 pages, open on Zenodo, DOI 10.5281/zenodo.23718472.
Second study: Trust, But Verify
A claim-level citation audit of AI-generated strategic research briefs — the companion study to the framework paper, open on Zenodo.
Check an AI vendor’s claims
A six-step check you can run on a vendor’s claims — using the vendor’s own materials — before you sign.
Check AI-generated citations
A four-verdict method for checking AI-generated citations, in about ten minutes per batch.
Policies
How this site handles data, and the terms for pilot engagements: privacy, terms of service, and refunds — pay only if the report passes.
Contact
Research collaboration, funding, or a pilot question — use the enquiry form, or write to the Centre directly. The form opens an email in your own mail app; nothing is sent or stored by this site.
Auditable by design.
Accountable by default.
We believe AI systems used in research and policy should be auditable by design: every material claim traceable to its evidence, uncertainty measured rather than hidden, and consequential judgements reserved for accountable humans. Our work is methodological and empirical — frameworks, evaluations and open publications.
Test the method with us.
Partnerships to pilot, score and improve auditable research workflows on real questions.
Discuss a collaboration ↗Back evidence, not hype.
Grant-funded research programmes on AI governance and evaluation for developing and emerging economies.
Discuss funding ↗Adopt AI with assurance.
Evaluation and assurance for institutions bringing agentic AI into research and policy work.
Discuss a use case ↗A modern Indian’s guide to civic sense & social etiquette.
How everyday actions — from traffic discipline to digital civility — quietly build a nation. The founder’s public-engagement book on civic behaviour and social responsibility in contemporary India.
The founders.
Amartya Kumar
Founded the Oryx Centre for Strategic Futures in Noida to work on the governance of advanced AI from a Global South vantage point. Author of the founding working paper and of INDIA: From Me to We (2025).
Riya Nalwa
Co-founded the Centre with Amartya and co-leads its research and partnerships — building the collaborations through which the framework will be piloted and tested.