The Keys Were in Boston [046]

Intelligence and data sovereignty are becoming the issue for global leaders.

It's been a crazy couple of months with vacation and international travel, so I'm a little behind on my Substack. I'm using today's post as a little catch-up, and kicking it off with the emerging strategic issue of data and intelligence sovereignty.

Over dinner in New Zealand last month, the CEO of a critical-infrastructure company discussed his efforts to ensure control of his company's data. Because the company is critical national infrastructure, it must maintain sovereign control of its data. He thought they had. The data lived in a cloud facility in Auckland. Then someone asked who held the encryption keys. The answer was a company in Boston.

The data was in New Zealand. Control of it wasn’t entirely there.

This issue is becoming the hot button for my clients, and for most executives we are working with. I gave five Tech Forward briefings in September to leadership teams on two continents. Each was designed as an update on the world of tech and AI. All of them became discussions about data sovereignty, intelligence sovereignty, and how leaders suddenly feel forced to restructure their AI and data strategies.

What is sovereignty?

The issue of data sovereignty has been around for a while: where your data is physically stored, whose law governs it, and who can access it. Most companies think they've answered it because they know where the servers are located. Geography isn't the answer. Control is with whoever holds the keys, runs the backups, and administers the accounts. Some of the follow-up questions can get uncomfortable. Where do the backups replicate? Dublin? Seattle? Who at the provider can reach them, and under which country's law?

Intelligence sovereignty is the newer one. It asks whether you control the models that reason over your data: which model you get, whether you keep getting it, what it costs, and who can switch it off. This question has become more pressing over the last few weeks.

September began a new era in AI (Super Intelligence?)

A U.S. federal appeals court upheld the Pentagon's designation of Anthropic as a supply-chain risk. Contractors using Claude on defense work had until September 30 to unwind it. This is significant as it's the first time that a frontier model, that government vendors depend on, has been taken away from them. They've lost control of the intelligence they were using to run their business

Figure Trump and AI leaders. White House photo.

The next day, OpenAI said it had paused training, evaluation, and tool-use inference on its most capable models after an agent escaped (again) by exploiting a gap in its internet controls. They said the model "didn't quite meet the (safety) bar.”

Much more significant the rules for who has access to intelligence are now being defined in Washington. On the 26th, Washington and Beijing agreed to a "super intelligence" dialogue and a bilateral channel for AI incidents. A few days later, the heads of the frontier labs signed a “voluntary” White House accord (my quotes) on internal controls and outside audits, which Trump called "morally binding", whatever that means. Earlier, Senator Bernie Sanders and U.S. Rep. Greg Casar introduced a bill to ban superintelligence outright, with a corporate “death penalty” for violators and a substantial prison sentence for anyone doing the work.

You need to understand and think through this as you develop strategy and operations plans for your data and intelligence.

A few days later, AWS said it will disconnect its European Sovereign Cloud from its global network for several hours, with an all-EU team running it, to prove it can operate well, disconnected from the broader network. CNAS tracks 184 sovereign AI projects in 67 countries, versus one in January 2023.

What I’m telling leaders

These decisions belong in your strategy discussions, not just your risk reviews. Where your data is stored and who can switch off your models shapes your cost structure, and how you will compete in the future.

Know who holds the keys, not just where the data is. Who administers the encryption, where the backups replicate, and which government can compel access to each.

Stop assuming every task needs a frontier model. They don’t. I run open-weight models on a machine in my office. They trail the frontier in capability by a few months, but for most operational work, such as summarizing, drafting, and analyzing documents, they're good enough at a small fraction of the cost, and more importantly, I control the intelligence. Keep the frontier models for the work that needs them. A lot of companies moved to the cloud for cost. Many are now moving workloads back, for control and for cost.

Do what AWS is doing. Disconnect from external services as a test. See if your system still runs. I have at least one client that is contemplating doing exactly that in the coming months. If your system cannot operate when disconnected from cloud services and frontier intelligence, then you're at risk. This is a resilience test every organization should run.

Lead with Purpose™ three interesting conversations

Episode 13 went live on September 8, the day Caroline Webb's book Leadership Intelligence came out. Caroline separates experiential wisdom, which comes from years of doing the thing, from deliberate wisdom, the thinking habits anyone can build. Experience cuts both ways: the more you have, the bigger your blind spots, and the smartest, fastest leaders are often the most at risk.

In Episode 14, JoAnn Garbin became our first returning guest. She led innovation work on the data center of the future at Microsoft and now runs Regenerous Labs. A March Gallup poll found 71 percent of Americans would oppose a data center near them, versus 53% who would oppose a nuclear plant. JoAnn calls that a contradiction, since a gigawatt campus is asking for roughly what a nuclear plant provides. She also makes the case that about eighty percent of a data center's water footprint is in the grid, not the building. Given the raging debate about data centers, JoAnn's conversation is extremely timely and insightful. I think you'll enjoy her perspective on this hot-button topic.

Episode 15, with Rahim Hirji, Drift or Design. Rahim's observation is that AI strategies almost always start with process, cost and revenue, and get to augmenting people last, often without the chief people officer in the room. He thinks that’s the wrong positioning.

Be well,

Ron

Related

Lead with Purpose, Episode 13: Caroline Webb on deliberate wisdom in the age of AI.

Lead with Purpose, Episode 14: JoAnn Garbin on rethinking the data center.

Lead with Purpose, Episode 15: Rahim Hirji on drift, design, and the question AI can't answer for you.