iNNOV8 has launched DARA — the Dynamic Analysis and Research Assistant — a human-supervised AI researcher based in Sulaymaniyah, Iraq. Its first published output, Between Knowledge and Algorithm: Generative AI in the Think Tank Environment, is an interesting read not simply because it is about AI and think tanks, but because it is an AI-generated, human-supervised research paper about the use of AI in think tanks.
That makes it more than another contribution to the growing debate on generative AI. It is also a practical experiment in disclosure, authorship, quality assurance and trust – topics that were at the centre of multiple discussions at the 2026 OTT Conference.
DARA is openly presented by iNNOV8 as a human-supervised AI research assistant. Its work is labelled as human-supervised and AI-generated. iNNOV8 says that DARA will propose, outline, select methodologies and draft research papers under the guidance of in-house researchers. It also commits to distinguishing DARA’s outputs from iNNOV8’s original human-authored research, and to making the methodology — and, where possible, the prompts — visible.
Many think tanks are already using AI. Researchers are using it to scan literature, draft outlines, summarise documents, translate text, test arguments and produce first drafts. Communications teams are using it to write social media posts, newsletters and summaries. Administrators are using it to organise workflows. Some leaders know this is happening. Others suspect it. In many cases, however, the practice is running ahead of the policy.
DARA makes the practice visible.
I think this visibility is the most important part of the initiative. It invites us to move beyond the tired question of whether think tanks should or should not use AI and ask under what conditions AI can be used in ways that strengthen, rather than weaken, trust?
From AI anxiety to trust infrastructure
At the 2026 OTT Conference in Rabat, our theme was Think Tanks and Trust. AI appeared again and again in the keynotes, parallel sessions and informal conversations. But it was rarely discussed as a purely technical issue. The conversation was institutional: what happens to think tanks’ credibility when machines can produce policy briefs, summaries, stakeholder maps and research drafts faster than any human team?
This was the point Erica Schoder made in her keynote. AI can do extraordinary “complexity work,” such as processing large quantities of information quickly and helping small organisations extend their capacity. But humans create meaning by deciding what matters, whose experience counts, what trade-offs are acceptable, and what consequences we are willing to own.
For think tanks, this distinction is crucial. Their value cannot rest only on the production of outputs. If that is all they offer, then AI will eventually do much of it faster and cheaper. Think tanks’ value also lies in judgment, relationships, accountability, legitimacy, care, political understanding and the capacity to bring people with different stakes into productive encounters.
DARA’s first paper is therefore very timely. It argues that think tanks need to separate three questions that are often blurred together:
- May AI be used?
- Must AI use be disclosed?
- Can the resulting output be trusted for publication?
That distinction is useful. A think tank may decide that AI can be used to transcribe an interview, but not to conduct it. It may allow AI to suggest a literature map, but not to replace direct engagement with the original sources. It may permit AI-generated first drafts, but only if a named human researcher can defend every analytical claim in the final output. It may use AI to identify patterns in large datasets, but not to make normative judgements on behalf of communities affected by policy decisions.
These are the everyday questions that think tanks face today.
We have been moving towards this point
OTT has been following the implications of AI for think tanks for several years.
In 2023, we published practical guidance on using ChatGPT in think tanks, with a simple but still relevant warning: always review what AI produces before using it. In 2024, I imagined AI assistants embedded in research, funding, strategy and policy engagement. Some of that future is already here.
Since then, the conversation has become more practical and more institutional. Aidan Muller argued that think tanks needed to move from opportunistic experimentation to preparedness. Joscha Wirtz, writing about Sociopúblico’s AI for Think Tanks cohort, framed the challenge as moving from FOMO to intentionality. Tony Bader argued that every think tank now needs an internal “AI constitution”: a clear, shared framework for what tools can be used, what data must never be uploaded, when disclosure is required, who is accountable, and how AI can support rather than replace critical thinking.
More recently, Andrea Cabrera and Kshipra Ajrekar have explored how think tanks should adapt to generative search. This is another dimension of the same issue. AI is not only changing how we produce knowledge but crucially how others find, interpret and judge it. In a world of AI-mediated search, think tanks need to make their work clearer, more structured, more discoverable and more trustworthy to both people and machines.
Trust in practice
iNNOV8 has chosen to name the assistant, define its role, label the output and explain the supervision behind it.
It is doing so to test the boundaries of our trust.
Readers are not simply being asked to trust the final text. They are being invited to understand the process behind it – which will not satisfy everyone. Some will argue that AI should not be used in research writing at all. Others will think iNNOV8 has been too cautious. But disagreement is precisely what the sector needs.
The first DARA paper also makes a useful proposal: a three-zone approach to AI use.
In the first zone, AI supports bounded or mechanical tasks where the researcher retains analytical authority.
In the second, AI makes a more substantive contribution and therefore requires stronger oversight and explicit disclosure.
In the third, AI should not substitute for human judgement, accountability or relationships — including primary data collection, participatory research, in-person convening and normative argumentation.
This resonates strongly with the discussions we had in Rabat. Again and again, participants returned to the idea that trust is not only produced by rigour. It is also produced by presence, the willingness and ability to listen, researcher proximity, care for ones’ stakeholders, transparency and accountability. AI may help us work faster, but it cannot take responsibility for what a recommendation means in a contested political context, build the trust required for a community to speak openly, nor replace the encounter between researchers, policymakers, funders, citizens and peers.
What think tanks can learn from DARA
I do not think every think tank needs its own DARA. But every think tank needs to answer the questions that DARA raises. And every think tank should pay attention.
First, think tanks need to know how AI is already being used inside their organisations. If staff are using AI informally, leaders need to create safe spaces to discuss the practice honestly.
Second, think tanks need internal rules that are simple enough to use. A long technical policy that nobody reads will not help. What is needed is a short, practical framework that clarifies permitted uses, prohibited uses, data protection rules, disclosure expectations and accountability.
Third, think tanks need to distinguish between assistance and substitution. AI can help with many tasks, but when it substitutes for the human work through which judgment is formed — reading, listening, debating, interviewing, revising, testing arguments, sitting with discomfort — it may produce thinner knowledge, even if the final output looks polished.
Fourth, think tanks need to make their trust practices visible. This means clearer methodology notes, disclosure of AI use where appropriate, stronger editorial processes, better metadata, transparent funding information and explicit quality assurance. Trust will increasingly depend not only on doing good work, but on making visible why that work deserves to be trusted.
Finally, funders should pay attention. Responsible AI use requires time, training, secure systems, editorial capacity and organisational learning. These are not overhead luxuries. They are part of the infrastructure of credible knowledge production.
A contribution from Sulaymaniyah to a global debate
It is also worth noting where this experiment comes from. DARA was not launched by a large institution in Washington, London, or Brussels. It comes from iNNOV8 in Sulaymaniyah. It suggests that the future of AI in the think tank sector should not be defined only by large, well-resourced organisations in the Global North.
Smaller and regional think tanks face particular risks in an AI-mediated information ecosystem. Their work may be less visible to large models, their languages may be less well represented, and their credibility signals may be harder for automated systems to read. At the same time, AI may offer them important opportunities to extend capacity, translate knowledge, reach wider audiences and participate more visibly in global debates.
The question is whether they can do so on their own terms.
DARA is one attempt to answer that question.
That is exactly the kind of experimentation the sector needs.