Washington, D.C.
2:00 p.m. ET
Remarks as Prepared for Delivery
Good afternoon. My name is Anjana Rajan, Co-Founder and CEO of Atalanta. Thank you all so much for being here today.
Before I begin, I’d like to take a moment to thank Ali Soufan and the entire team at The Soufan Center. This report, titled “Closing the Gap: Software Understanding and U.S. National Security”, is serious, rigorous work, and is exactly the kind of analysis this moment demands. We are proud to have supported it, and deeply grateful to be part of this conversation.
Ten days ago, something remarkable happened that perfectly frames why we’re all here today.
Anthropic built one of the most capable AI models ever developed. The U.S. government couldn’t credibly understand what it would do in the wild – what vulnerabilities it might expose, or what it might enable in the wrong hands. So the government made a call: restrict access unilaterally. They placed export controls on Fable, Anthropic’s most advanced model, effectively overnight. Anthropic disabled access for all customers to comply.
A debate erupted across government and industry. And at the center of it was a question neither side had the tools to fully answer: is this software-defined system safe to deploy?
This is not primarily an AI governance story. It’s a software understanding story. And it’s exactly the story The Soufan Center’s report tells.
We are constantly presented with false choices when it comes to technology. Speed or safety. Deploy or restrict. Innovate or regulate.
What we watched last week is what happens when that false choice becomes unavoidable. When a system is powerful enough that no one can be certain what it will do, the only tools available are a jailbreak patch or an export ban. Industry pushes for speed. The government pumps the brakes. Both sides are trying to respond to a problem neither has the tools to actually solve.
When you cannot understand a system, restricting it can seem like a reasonable response. But it is not a sustainable strategy. And it cannot be the ceiling on how this country governs its most powerful technologies.
Software understanding offers a third path – one where we don’t have to choose between capability and safety because we have built the tools to verify both. Where vulnerabilities are found and eliminated during development, not discovered after deployment. Where the question of whether a system is safe to deploy has rigorous, mathematical answers, not ones improvised under pressure.
That is what this report is about. And that is what Atalanta was built to do.
Atalanta is a mathematical AI company. We help the U.S. government and the American industrial base design, develop, and deploy critical systems — faster, cheaper, and with higher assurance. Our product, Argo, combines three things: AI for speed, formal methods for rigor, and digital engineering for systems thinking. Together, they make something possible that hasn't been before: provably correct decision-making for the world's most important missions.
The report you’re holding today documents six ways the software understanding gap is actively eroding U.S. national security. The names are familiar: Salt Typhoon. SolarWinds. Volt Typhoon. NotPetya. Colonial Pipeline.
What unites them is a single structural failure: we build and deploy software faster than we can understand it. That gap between what we deploy and what we can actually reason about is the terrain our adversaries have learned to inhabit.
China’s Volt Typhoon is already pre-positioned inside U.S. utilities and transportation hubs. Not causing disruption today, but waiting patiently. Ready to act, at the moment of their choosing, in a Taiwan contingency when we are trying to mobilize. Russia’s SolarWinds intrusion sat undetected for nearly a year. The FBI has said Salt Typhoon remains ongoing.
These are failures of software understanding. You cannot defend what you cannot comprehend. And now, we are deploying AI systems that will make consequential decisions autonomously, at speeds no human can supervise, inside the same infrastructure that is already compromised. What happened with Anthropic this month is not an anomaly. It’s a preview. Every frontier model release from here forward will involve some version of this same question: do we actually know what this system will do?
Right now the honest answer is: not well enough.
Here is where I want to give you reason for optimism.
Recently, Atalanta announced a partnership with Idaho National Laboratory in support of the Genesis Mission – a national AI initiative led by the White House and the Department of Energy to harness artificial intelligence for scientific discovery. We are deploying Argo – the first software understanding platform – to prove that AI systems operating inside critical infrastructure perform correctly under real-world conditions.
This is exactly the problem at the heart of the Anthropic debate. How do you trust an AI system in a high-consequence environment? Restricting access is one answer. Dismissing the seriousness of the problem is another. But Software Understanding is undeniably the best one.
Traditional testing shows you how a system behaves in scenarios you’ve already imagined. Argo goes further. It systematically checks every possible state a system can reach and reasons about entire classes of inputs – including adversarial ones – to prove that critical properties hold.
Congress agrees. Through the 2026 National Defense Authorization Act, the Senate Armed Services Committee has directed the Pentagon to develop a comprehensive strategy for transitioning formal methods into production environments at scale. The pieces are converging: the technology is ready, the government is directing it, and the missions are demanding it. What’s needed is the will, across this entire ecosystem, to close the gap.
The debate we watched this month will happen again. There will be more models, more disputes, more emergency decisions made under pressure with inadequate tools. That cycle does not end until we build the capability to actually understand these systems.
I can think of no one better to explore what that requires than the people you’ll be hearing from today.
The Honorable Chris Inglis served as our nation’s first National Cyber Director. He has lived this problem from the inside, at the highest levels of government, and he is here because he believes – as do I – that this is solvable.
Following that, you will hear from the authors of the report. Dr. Colin Clarke and Clara Broekaert from The Soufan Center will walk you through the research. I will join them to offer a technical perspective on closing the gap.
Both of these discussions will be moderated by the incomparable Steve Clemons, who will help connect the dots between technology and national security policy and what this moment demands of all of us.
At its core, closing the software understanding gap restores human control over human creation and tilts us toward a software-defined world worthy of our trust, resilient against those who would seek to undermine it, and aligned with the future we want to live in.
This is a conversation we need to be having. I’m glad we’re having it here today, together.
Thank you.