She and Isbister co-chair the closed-door dialogue, and selection for it is made jointly by the judges, the Lab and DFAT. She is the single person most able to change what this day is worth to you.

"Geordie's Lowy piece in July made the sovereign compute case better than most of the industry has. We're the small end of that argument — the version that doesn't need a data centre at all."

Ex-Attorney-General's Department, EY senior manager, ACFID director, anti-corruption adviser embedded in PNG's Department of Justice. Fulbright scholar. Non-resident affiliate at CSIS Washington and ANU RegNet. The Lab is a five-person private company, not an institution — which is exactly why access runs through her personally.

A place in the closed-door dialogue, and her read on where DFAT's AI thinking is actually heading.

Co-chairs the closed-door dialogue. His career spine is humanitarian response, then Ambassador for the Environment, then Deputy Head of the Office of the Pacific and Head of Strategic Infrastructure. He thinks in deployable regional infrastructure, not pilots. That is unusually good news for a product that is literally a physical thing you install.

"You asked what Indo-Pacific development challenge AI could solve by 2031. Ours is the one where the answer can't be a data centre."

What DFAT would need to see before local, owned AI becomes procurable inside a development programme rather than an interesting exhibit.

Her portfolios intersect exactly on you. An Australian small business selling sovereign AI into the development programme is literally two of her three jobs at once. Almost nobody in that room will say that to her out loud.

"You've argued aid transparency has to become meaningful, not just technical. We build the machine so the organisation can actually answer how a decision was made — because they own it."

Her signature theme in this portfolio is transparency as a strategic asset against disinformation (Devpolicy, August 2025). Her academic career is on extremism and media discourse, so frame provenance and trust, not compliance. She took International Development from Pat Conroy in May 2025.

He published the sovereign compute argument in the Lowy Interpreter on 14 July 2026. He is the most aligned thinker in the building and he works for your host. If one person there should walk away evangelising Selma internally, it is him.

"Your line about not dressing the Pacific data centre tension up in development language — that's the reason we build a box instead of a cloud."

His piece cites Anthropic seeking 1.4 gigawatts of Australian data centre capacity, up to US$15bn, and notes 64% of Australians think AI's risks outweigh its benefits. He frames compute as negotiating leverage, not just capability.

For him to see Selma as the physical instance of his own published argument, and to say so to Bridi.

Their published AI Principles for Development commit to equitable access covering "distribution but also ownership, maintenance, and support", and to low-income countries as "essential collaborators", not end-users. Selma sits exactly on the seam between those principles and their flagship AI play.

"Your AI principles talk about ownership, maintenance and support, not just distribution. That's a hardware sentence more than a software one."

In May 2026 they announced a $200m, four-year partnership with Anthropic across health, education and agriculture. No Australian or Pacific office — Singapore is the Asia hub, opened 2026, and its published scope covers South and Southeast Asia with no mention of the Pacific. Sri Mulyani Indrawati joined the board in January 2026.

The highest-priority commercial conversation of the day. A US development giant buying Australian market position as American firms pivot here after the USAID cuts. They are shortlisted on current DFAT tenders including an Indonesian economic governance programme worth over half a billion dollars across ten years. A US firm bidding Australian work needs Australian credibility and a differentiated technical offer. That is precisely what you are.

"You're bidding Australian work as a US firm. What's the sovereignty question you keep getting asked, and what do you currently answer?"

To be the named sovereign AI component in a live bid. Not a partnership in principle — a specific tender.

Canberra office, 50 years old, 160+ countries. They describe themselves as working at "the intersection of technology, data, and human progress", which is an opening you can use verbatim.

The largest DFAT contractor by a wide margin — roughly $2.8bn in annual average terms, ahead of Abt and Palladium. The Frontier Tech Hub is FCDO-funded and exists to pilot frontier technology in development contexts. It is a ready-made procurement vehicle for exactly this.

"You're showcasing too, so peer to peer — what do your AI pilots keep failing on? I'll bet it's connectivity or data governance."

Selma as a testable frontier-tech pilot with a named programme behind it.

A small Australian outfit that has been shipping real software into Timor-Leste, PNG, Solomon Islands and Fiji for nearly two decades. Their maternal health service Liga Inan doubled skilled birth deliveries in Timor-Leste. They have already solved the offline-first, local-ownership, government-adoption problems that will sink most of the pitches in that room.

"What breaks first when you deploy into a provincial facility? I want the unglamorous answer."

Deployment reality, honestly given. And possibly a partner who could install and support Selma units in-country, which is a problem you have not solved outside Australia.

DFAT-funded platform connecting Indonesian and Australian research and innovation. Not a buyer — a funding and access route into Indonesian institutions, where data localisation is a live and hardening policy constraint.

"Indonesia's PDP Law and GR 71 make public-sector data placement a real constraint. What are your partners doing about AI on restricted data?"

An introduction to an Indonesian institution with a data placement problem and a research budget.

Real distribution into community-sector not-for-profits across Asia-Pacific, and a track record of moving philanthropic technology money — $1.5m from Google.org in 2024 to help regional non-profits scale. Not a big-ticket buyer. A genuine channel to many small ones, which is the shape of your actual market.

"Your Digital Capabilities Program orgs — how many of them are quietly putting client records into a free chatbot right now?"

Spriggs also sits on the Australian AI CoLab steering committee alongside Treasury, NSW Government and ANZSOG. That is a line into domestic public sector AI adoption, which may be the more addressable market than the Indo-Pacific one.

The strongest philosophical match in the room. They are adapting Trust Intelligence as a monitoring and evaluation tool for First Nations engagement in international development. Indigenous Data Sovereignty is the one domain where "this data must not leave the community's control" is a stated ethical principle rather than a compliance preference — which is Selma's entire architecture, expressed as an ethic.

"Where does the trust data physically sit, and who decides that? Not the policy answer — the actual server."

Ashlee Wone is a Ngiyampaa Wangaaypuwan, Wiradjuri and Ngemba woman, founder and Principal of Wone Consulting, based in Western Sydney. Trust Intelligence measures the gap between intended and experienced trust.

He will understand Selma faster than anyone else in the building. Eighteen years in Pacific, Asian and African health systems. Won the 2018 Public Sector Innovation Award for Tupaia. His pitch: Pacific disease detection currently relies on weekly manually-reviewed syndromic surveillance with response times over seven days.

"Under the IHR, when to declare an outbreak is a sovereign decision with trade and tourism consequences. Does a health ministry really want that signal appearing on foreign infrastructure before its own minister has seen it?"

Re-identification: in Pacific populations of tens of thousands, "three cases in this district this week" is effectively identifiable. Sovereignty of the signal: the IHR point above, which has nothing to do with privacy law and is unusually persuasive. Connectivity: Tupaia's entire design premise is intermittent, expensive connectivity in Pacific health facilities.

An honest answer on whether your hardware assumptions survive contact with a provincial Pacific health facility. He is the one person who can tell you today.

Commercially the most viable conversation on the floor. They run AI transcription, translation and synthesis over recorded voice and video testimony from programme beneficiaries across the Indo-Pacific. Clients include the World Health Organization and the World Bank. That is identifiable testimony from vulnerable people, moving through a cloud.

"Two problems you must have: where identifiable testimony is allowed to sit, and what happens when there's no bandwidth. A local transcription tier solves both without touching your platform."

A conversation about Selma as a local inference tier alongside their cloud product, not a replacement for it.

Small team. Canberra-headquartered per the booklet, though Crunchbase says Brisbane. David Lloyd-Lewis is the other co-founder and Chief Product Officer.

The only pitch with a confirmed live deployment. Announced March 2026 with Surgo Health, a US public benefit corporation: multilingual AI voice and chat collecting lived-experience data from Barangay Health Workers in the Philippines, scaling to around a thousand workers, feeding a government-facing dashboard. The release cites consent protocols and says nothing about where the data is hosted or processed.

"Not your pilot — the government side. When the Department of Health wants that dashboard, are they comfortable with the analytics layer sitting offshore?"

Morgana Ryan is not a technology-naive NGO chief executive. She led Accenture Development Partnerships in Asia Pacific and chaired the Infoxchange board. CARE's HEAL Hub in the Philippines is already designed to work offline — they build for connectivity gaps by default.

The textbook indigenous data sovereignty case: recordings of elders, legends, traditional songs, clan-restricted material, across PNG's 800+ languages. In his own words, the younger generation lose their languages while elders pass on with the knowledge. This is the case where communities genuinely refuse cloud processing.

"Whose permission does a recording of an elder need before it goes anywhere? That question is the whole design brief, and most AI companies never ask it."

From Lufa, Eastern Highlands Province. Maths and computer science, a decade in oil and gas, then business management. Timing matters: PNG launched Kumul Cloud Infinity, its first sovereign AI data centre, on 26 August 2026 — three weeks before this event. Telikom's CEO said "almost 100 percent of our data has been managed and stored abroad".

Your best ideological ally, and the person most likely to tell you how Selma should be framed for Pacific audiences. Publicly on the record about indigenous knowledge informing AI policy, Global South representation in AI governance, and — directly relevant — the environmental impact of AI infrastructure on island nations.

"You've written about what AI infrastructure costs island nations environmentally. Ours runs off a power point and doesn't need a cooling tower. Does that argument land in Palau, or does it sound like marketing?"

PIF foreign ministers endorsed Pacific Digital and AI Technology Principles in August 2026, going to Leaders in Palau, with a regional AI Governance Framework in development for 2027. She is Palauan and works on exactly this.

Australian lawyer, seven years across the Indo-Pacific with governments and international institutions. Recent visits to Tuvalu and Kiribati prompted her to ask whether AI can help Pacific states preserve the legal lines that define a nation as seas rise. Maritime baselines are literally sovereignty-defining data — they set the outer limit of an EEZ and therefore fisheries and seabed rights.

"Baselines under UNCLOS, or a baseline survey of assets? They're very different problems and only one of them is about sovereignty."

The PIF's 2021 Declaration on Preserving Maritime Zones commits members to fixing baselines and depositing charts with the UN so EEZs don't shrink with sea level rise. The 2025 ICJ climate advisory opinion has sharpened it. The underlying hydrographic data is nationally controlled and sometimes defence-classified.

Near-perfect data profile. A criminology and justice institute working on climate-driven exploitation risk across India, Nepal and Sri Lanka, gathering community-reported information about vulnerability. Identifiable informants plus trafficking and exploitation risk is among the most protection-sensitive data in development work. They publish no data handling policy, which is itself notable.

"You're holding community-reported exploitation risk data. Where does it go when you analyse it?"

Technically your natural habitat. Drone imagery and computer vision for rapid post-disaster damage assessment, with communities training their own models through HOT's open-source fAIr platform. Disaster zones with no connectivity, local model training on local data. That is Selma's physical premise, arrived at independently.

"Community-trained models on local data is the same instinct we had, just at a different layer. Where do you actually run the training when the network's down?"

They are the Lab's existing AI partner and have been since 2024. Bridi Rice is publicly on record that Lab analysts produced solid Southeast Asian policy analysis in two hours using their tool. They are a collaborator, a showcaser, and named on the Lab's own AI page. And their marketing already says they run "in client environments (cloud or sovereign), no vendor lock-in".

"You're an analysis layer, we're the substrate. What happens when a client's material is too classified to leave the building at all?"

They are a method and a workflow, not an appliance. Their reasoning workflows could plausibly run on Selma for classified policy work. Go in as a possible substrate, not a rival.

Dragonfly and the Lab have jointly published concerns about Western knowledge dominance, donor assumptions displacing local realities, energy consumption, and the need for human oversight. That is the Lab's house view on AI risk, and it is the frame your showcase will be judged against. Human at the Helm answers three of the four directly.

They already market "isolated private instances accessible only to your organisation", no training on customer data, no external sharing. That blunts your differentiation with the same probity-sensitive government buyers.

"An isolated private instance is still someone else's cloud, in someone else's jurisdiction, under their retention terms. Ours is a machine in your building with nothing plugged into it."

They sell to rūnanga and tribal councils — the same Indigenous data sovereignty territory as Empathos and Wone. Could go either way. Go in curious rather than selling.

Their model is collecting and processing "hyperlocal data from hard-to-access areas of the world" — which is, precisely, offshore processing of other people's countries' data. They are the philosophical counterpoint in the room.

"They take the data out to understand a place. We leave the data where it is and take the machine in. Same sector, opposite architecture."

CEO Erol Yayboke is ex-CSIS — and Bridi Rice is a non-resident CSIS affiliate. Assume warmth between them. Useful to understand, risky to be conflated with.

Not a prospect. They hold exactly your data class — live patient consultations — and have solved it the opposite way, with certifications and cloud scale. Valued around US$465m. They matter as the incumbent answer a health buyer will reach for: "why not just use Heidi?"

"Certification is a promise about how someone else handles your data. Ours is a physical fact about where your data is. Plenty of organisations can accept the first. Ours are the ones that can't."

Chase them as a customer. At their funding level they would build any on-premise tier themselves.

An AI assurance platform that tests draft reports against their own evidence base. Cloud-only, single AWS London region, no on-premise option. That is a live constraint for any government client who can't send draft evaluations offshore, and it is a gap you could fill.

"Your whole product is checking claims against evidence, which is our Trace argument in a different shape. What do you tell a client who can't let the draft leave the country?"

A UK-jurisdiction cloud service pitching Australian government evaluators is the argument you're making, running in the other direction. Useful, and worth being gracious about rather than smug.

Adimi — an AI education team building a creativity-focused learning app for young learners, with AI-supported lessons and guided drawing that "preserve learner agency". Children's data by definition, which is a regulated category. Claim — a free tool turning a community's plain description of a problem into a funder-ready proposal, plus grant eligibility matching. Two co-founders in the environmental and international development space.

Neither has an obvious sovereignty problem. Low priority unless the conversation surprises you.

Small Island Developing States lack both infrastructure and capacity. Most are still building basic ICT and are nowhere near looking at AI infrastructure. Her line: "For now, we only have our data, our unique Pacific values and traditional knowledge, and our combined Blue Pacific Continent." The conclusion she draws is not procurement, it is a seat at the table, approached collectively as a region.

She names the barrier as infrastructure, and the thing she is picturing is a data centre. Selma genuinely removes that one. One machine, no facility, no cooling towers, no water, and it keeps working when the cable doesn't. That is a real change in order of magnitude, not a marketing angle.

But read the whole sentence. She said infrastructure and capacity, and capacity is people, institutions, governance and someone to run the thing. A box supplies none of it. She also says they are still building basic ICT, which is the layer underneath Selma, not the layer Selma replaces.

The energy case — she calls AI the most energy-hungry technology ever built — and technology-facilitated gender-based violence, which rose in Palau after the fibre cable landed in 2017. Her framing is "both the everyday and existential risks". The everyday half is the half most vendors skip.

Warns against external systems built on assumptions that overlook Pacific languages and cultures, citing AI-generated Polynesian art and tattoos as the visible edge of it. Proposes a Regional AI Governance Framework, joint AI toolkits, shared digital public infrastructure and a Pacific Islands AI Technical Assistance Facility. Explicitly criticises short-term, fly-in fly-out projects. Her close: AI sovereignty as a pathway to self-determination that "starts with investing in Pacific expertise".

She has written the takedown of a badly pitched Selma in advance. An Australian vendor arriving at a Canberra conference to explain what the Pacific needs is the exact silhouette she is describing. That is not a reason to avoid her position; it is a reason to know its shape before someone else names it out loud at the stand.

The honest counter is architectural, not rhetorical. Selma is configured and grounded in what the organisation puts into it and answers only from that, so the assumptions in the room stay the organisation's own. Offer that as a property of the machine, never as a claim about culture.

"You're right that most of this arrives as someone else's assumptions. The only honest answer I have is architectural: it runs on what you put into it and nothing else, and it stays in your building. What that's worth is yours to judge, not mine."

Splits AI sovereignty into two elements: data access and compute power, then sovereignty and the rule of law. Proposes a shared Pacific AI Digital Hub alongside Digital Embassy Facilities. His close: "For the Pacific, AI sovereignty is about preservation of the Pacific inheritance and way of life."

The closest of the three to Selma's actual shape, and the most useful person to think with. His first element is what a Selma unit is. His second element is where our gap sits, and it is worth being straight about it: the data stays in country, but the product is Australian, the model underneath was built offshore by someone else, and the client leases rather than owns outright. That is a defensible arrangement. It is not the same as the Pacific owning its AI infrastructure, and pretending otherwise is how you lose a technical audience.

"Where does a physical unit sit in your two elements? It answers the compute half outright. I'm less sure it answers the rule-of-law half, and I'd rather hear that from you than guess."

Ngoriakl lists traditional knowledge as one of the three things the region actually holds. Lagakali is describing what happens when it gets processed by systems built elsewhere. It is a live and contested area with its own frameworks behind it, and it is the single most tempting thing to overclaim on, because local and closed genuinely is a better answer than anything cloud-based.

The true framing is not "we solved this". It is: the thing you already have is the thing this runs on. Their data, their material, their values, in their building, nothing leaving. That is accurate, it is respectful, and it lands on exactly the asset they are protecting.