This spring’s summit ran the full gamut: US fiscal policy, labor markets, defense spending, national security. During Tuesday’s dinner at the Capitol, a long-time member of the House of Representatives with a track record of cross-aisle work provided candid comments. On Wednesday, members of Congress, Presidents of US Universities, CEOs and members of the organization, and a host of economic and policy folks from the US and EU governments all spoke on topics of key concern to America’s largest companies. It was a long day of discussing and learning and it was worth every minute.

These gatherings operate under Chatham House rules so my perspectives here will necessarily not identify people in the room, I want to share some insights I personally gleaned from the event. Not the headline geopolitical stuff (you can get that anywhere) but the conversations that highlighted what most organisations are quietly (and sometimes not so quietly) struggling with right now. The through-line, whether we were talking about AI adoption or workforce strategy or decision-making under uncertainty, kept coming back to the same thing: trust.

The AI Problem Is a Human Training Problem

One of the most clarifying moments of the day came during sidebar discussions about where AI adoption is actually breaking down inside organizations. The consensus is that the hardest part isn’t the technology.

The hardest part is in making the AI work for the humans and what they need to adapt. That is why ERI brought in intelligence operator, briefer and trainer extraordinaire Dr. David Priess a year ago to spearhead our own workforce transformation.

You can’t just mandate AI use and expect transformation. What’s becoming clear is that deploying AI effectively requires organizations to fundamentally re-order the way their people work, the workflows, the decision loops, the habits. That’s much more than just adding a new piece of software into the mix. It’s organizational change management at its most demanding.

This resonated deeply with what we see at ERI every day. The risk and intelligence teams that are getting real value out of AI are the ones that have slowed down enough to ask: how does this actually fit into how my team thinks and works?

They are dissecting job descriptions into human and agent roles and re-defining what that means for every job description in the organization. The ones struggling most are the ones who handed out licenses and waited for magic to happen. There, trust is already on the bubble as employees try to gauge whether the boss wants them to use it to augment their work, or to replace it.

AI in the Decision Chain

A related problem that we’ve been contemplating that came up repeatedly is a decision intelligence and compliance problem. Despite the growing popularity of tools like Claude that can spit out a polished and impressive looking intelligence report but with no guarantee of veracity and information origin – accountability, accuracy and audit trails are becoming a significant concern. Regardless of how polished AI makes an assessment look and feel, decisionmakers are still required to demonstrate a defensible audit trail for information used in consequential decisions when things go wrong and legal teams get involved. For risk management and security teams, but also executive leadership and boards this may be the difference between a major insurance payout, a ruling in a legal dispute or worse.

If your organization makes decisions that affect people – at its most crucial, their safety and survival alongside business continuity – including contributing to force majeure decisions you could eventually face a board or even a courtroom to discuss and defend how a decision was made. AI does not yet provide us with that audit trail for investigation and legal discovery post a major incident.

Last week’s discussions really solidified for me that if you are accountable for one of these functions then, in the very near future, the question of - ‘how do you know that information was accurate, where did it come from and who signed off on it?’ is coming for you.

This is the same logic that gave us financial audit requirements, environmental compliance frameworks, and clinical trial documentation standards. Complex systems that affect people require accountability structures. AI is a complex system that affects people.

Large language models like the ones powering most enterprise AI tools right now are extraordinarily capable. They are also trained on the open internet, which includes state-sponsored propaganda, disinformation, deeply embedded misconceptions, and narratives that bad actors have spent years and significant resources pushing into the information ecosystem. And yes, they still hallucinate. That is not a solved problem.

This is certainly not a reason to avoid AI. It is a very good reason to train your workforce and build structured, repeatable approaches that can assess what your AI tools are actually telling you, check the veracity of where that information comes from, and build trust into the system.

At ERI, we have been doing this work for human-generated intelligence for decades. The principles now are the same, but the stakes are higher, because AI scales everything. Including errors.

The Workforce Split

The last thing I’ll share is that companies are also going to have to spend time on building trust in AI from within their own workforce.

We are watching a real and growing divide emerge between workers - largely Gen X and late Millennials - in their relationship with AI. Some have embraced it deeply. Others have actively resisted it or use it minimally and with significant skepticism. Neither position is irrational given where we are.

But. The consequence of this is that two people with identical degrees, identical titles, and identical years of experience are going to show up to the same job with dramatically different skillsets, knowledge, efficiency, and quality of output, based almost entirely on their AI fluency. The same applies to existing staff, who will likely not apply AI consistently.

That is a practical challenge that very few organizations have started to grapple with seriously. And it compounds quickly, because the people who are using AI well are moving faster, doing more, and frankly becoming more valuable. Not because AI replaced their expertise, but because it freed them to apply it more fully.

I came away from Washington with a lot to chew on, and more than a few new conversations I want to have with clients and partners in the coming weeks. If any of this resonates with something you’re working through in your own organization, I’d genuinely love to talk.

As always: the risk isn’t out there somewhere in the abstract. It’s in the decisions being made right now, in organizations just like yours, by people trying to navigate something genuinely new. The best thing we can do is face it clearly, with good information and good judgment.

That’s still the job. It’s just gotten a lot more interesting.