Inside 'Dialog': What a Leaked Elite Network Reveals About AI Governance and Conflicts of Interest

Leaked ‘Dialog’ network puts AI governance and conflicts of interest under the spotlight

A Reddit post this week highlights a verified leak of “Dialog”, a private society founded in 2006 with no public website, undisclosed members, and a strong culture of confidentiality. According to the post, WIRED has verified a membership list and agenda for an August retreat in Dublin attended by 222 people. The list reportedly includes senior US officials (Treasury, Army), a senator overseeing the FTC, a NATO commander, the co-founder of Palantir, and OpenAI’s Chief Strategy Officer.

The AI angle is straightforward: the people building the models, the people regulating them, the people funding them, and the people distributing them were all in the same room – with no public record of what was discussed. That is not necessarily illegal, but it raises predictable questions about transparency, influence, and accountability.

Original Reddit thread: The people building AI and the people regulating it have been meeting in secret for 20 years.

What is “Dialog” and why are people concerned?

Per the Reddit post, Dialog was co-founded by Peter Thiel and Auren Hoffman in 2006. There is no public membership list and no official site. Attendees to the recent retreat reportedly registered with personal or corporate emails rather than .gov addresses – which the poster argues keeps communications outside the scope of US Freedom of Information Act (FOIA) requests.

Session titles listed include “Navigating WWIII”, “Battlefield Technologies”, and “Build-a-Cult”. The post’s author has published an archive of the verified membership data and documented conflicts of interest here: build-a-cult.com. They frame it as a research tool rather than a hit piece, with sources named and linked.

“It’s not a conspiracy theory. It’s a structural problem.”

Key terms and why they matter

• AI governance – the policies, processes, and oversight shaping how AI is developed, deployed, and audited.
• Regulatory capture – when industries heavily influence the regulators meant to oversee them.
• FOIA/FOI – laws that allow the public to request information from government bodies. Use of private emails can reduce visibility of official business.

Where AI meets power: builders, funders, regulators in one room

Private off-the-record meetings between industry and government are not unusual. There are benefits: candid exchange, faster crisis coordination, and better cross-domain understanding. However, AI is rapidly becoming a general-purpose technology shaping security, infrastructure, finance, and public services. When the people with the greatest commercial stakes sit with the people writing or enforcing the rules – without minutes, disclosures, or public scrutiny – the risks of conflicts of interest grow.

For AI specifically, the overlap runs wide: cloud platforms distributing models, venture funds backing labs, labs setting safety narratives, and policymakers grappling with limited technical capacity. Informal networks can set de facto standards and political direction well before formal consultation ever happens.

FOIA, private emails, and the UK context

The Reddit post notes that US officials used personal or corporate emails to register for the Dialog event, which would generally keep related correspondence out of FOIA scope. In the UK, the Freedom of Information Act 2000 and guidance from the Information Commissioner’s Office (ICO) make clear that information relating to official business can still be subject to FOI even if created or stored on private accounts or WhatsApp.

In practice, however, retrieval depends on records management, cooperation, and culture. If sensitive policymaking moves to private channels and off-the-record gatherings, FOI rights become much harder to exercise. That weakens public trust and the quality of scrutiny.

Implications for UK developers, enterprises, and policymakers

Why this matters beyond Washington

• Market direction – If core norms and roadmaps for AI safety, security, and access are shaped privately, UK-facing standards and vendor offerings may reflect those choices by default.
• Regulatory framing – Narratives around “frontier risk”, open vs closed models, and safety requirements can shift procurement, liability, and compliance costs here too.
• Public interest – Sensitive topics like battlefield technologies and dual-use capabilities affect allied policy, export controls, and academic collaboration.

Balanced view: potential benefits and risks of private dialogue

• Benefits: frank security briefings, relationship-building across silos, rapid coordination during crises, and early identification of cross-border risks.
• Risks: regulatory capture, unrecorded lobbying, unfair market advantages, and policy drift away from public-interest outcomes and competition.

Practical steps for UK organisations building or buying AI

• Ask vendors governance questions – Who funds you? What boards or advisory roles do leaders hold? What safety commitments are public and auditable?
• Demand disclosure in procurement – Include conflict-of-interest declarations, model documentation, data lineage, and an incident response plan.
• Keep your own audit trail – Record decisions, risk assessments, and model changes. For public-sector work, ensure FOI-ready record keeping, even across private channels.
• Do a Data Protection Impact Assessment (DPIA) – Especially where personal or sensitive data is processed, or where model outputs inform consequential decisions.
• Balance vendor claims with external evidence – Prefer primary sources like model cards, pricing pages, and independent evaluations over marketing.

What to watch next

• Transparency moves – Will any attendees disclose participation or outcomes? Are agendas, minutes, or conflicts registers published elsewhere? Not disclosed at time of writing.
• Spillover into standards and policy – Look for alignment between private forum talking points and subsequent regulatory or industry guidance on AI safety and access.
• Infrastructure externalities – Closed-door decisions often shape compute, energy, and water demand. For a deeper look at one of these hidden costs, see: AI data centres and water use – what the numbers really mean.

How to engage constructively

Secrecy breeds suspicion, but the solution is better governance, not outrage. If you work in or with government, push for:

• Clear rules on meetings – Publish attendee lists, agendas, and summaries where possible. Record and retain communications even on private channels when conducting official business.
• Conflicts registers – Publicly log external roles and interests for senior officials and advisers working on AI policy.
• Open consultation – Engage widely with academia, SMEs, civil society, and international partners beyond the usual Big Tech circles.

Final take

The Dialog leak, as presented in the Reddit post and associated archive, is a reminder that AI governance is not just about models and benchmarks. It is about who is in the room, what they are incentivised to do, and whether the public can see and challenge the direction of travel.

Private dialogue has its place, especially on security. But if the future of AI is shaped primarily in spaces without records or accountability, we should expect policy that serves the already-powerful. The fix is not to shut doors, but to open windows: publish more, disclose more, and design processes that withstand sunlight.