I was thinking about that reality when I sat down to judge the Financial and Cyber Risk Intelligence track at 2026’s Innovate DTCC: Industry-Powered AI Hackathon, supported by FINOS. It framed everything I evaluated. Not whether a solution was technically impressive in isolation, but whether it would actually survive contact with that environment.

What I found across five days of judging surprised me, in the best way possible.

Beyond the Demo: Designing for the Operating Environment

The teams in this track did not show up with polished solutions to invented problems. They engaged with the real friction: the fragmentation of security data, the cognitive overload on analysts, the challenge of making AI outputs legible enough for a risk officer who needs to act in thirty minutes on incomplete information. That level of problem clarity is rarer than it sounds. Most innovation work in this space optimizes for demo performance. These teams were thinking about what happens after the demo.

The strongest submissions shared a discipline I kept coming back to: they had chosen a specific failure in the operating environment and designed backwards from it. Not "AI for cyber risk" as a category, but a particular gap, a particular user, a particular breakdown that nobody had yet solved cleanly. That specificity was what separated work that felt production-ready from work that felt theoretical.

Explainability came up repeatedly as a real design constraint, and rightly so. A model that flags an anomaly has done roughly ten percent of the job. The other ninety percent is giving the analyst, the auditor, the risk officer, enough context to act with confidence. Regulatory environments in financial services do not accommodate black boxes, especially in risk functions. Teams that treated interpretability as a structural requirement from the beginning rather than a feature bolted on at the end understood something important about where this technology actually has to operate.

Adversarial robustness was the harder conversation. Cyber actors adapt. They study detection patterns and route around them. A model trained on historical attack vectors performs well on historical attack vectors, and not necessarily on what happens next week. The teams that built in monitoring, drift detection, or some form of feedback loop to account for that reality earned considerable respect from me regardless of where they landed on other dimensions. Acknowledging the limits of your system is itself a form of sophistication.

The Gold Award

Now to the Gold Award, because that team deserves to be named directly.

The "Snowflake and DTCC: Improving Security Posture with AI" team went straight at the problem I opened with. The six-hour investigation. The fragmented dashboards. The analyst burnout. Rather than making that process incrementally faster, they reimagined it entirely.

What they built is an AI-powered, natural-language threat hunting platform that unifies seven major security data sources into a single conversational interface. No complex SIEM queries. No context switching between tools. Instead, a multi-turn, context-aware investigation experience powered by Snowflake Cortex Agents, capable of generating AI-driven threat descriptions and risk assessments that compress hours of manual analysis into minutes.

What made this the Gold winner was not just the ambition. It was the execution. The architecture was scalable. The live demo was genuinely compelling. The solution was accessible enough for analysts without deep coding experience, which matters in a field already stretched thin on talent. The collaboration between DTCC and Snowflake was visible throughout in the coherence of what they presented. These were not two organizations showing up with separate priorities. They had actually built something together.

The business case was also unusually clear for a hackathon submission. Faster threat response, improved security posture, threat hunting capabilities no longer gatekept by technical depth. The judges knew quickly that we were looking at something with a credible path beyond the demo stage.

Open Source as Shared Infrastructure

I also want to say something about the format, because the choices DTCC and FINOS made in how they structured this event are not incidental to the quality of what was produced.

Providing Cloud infrastructure. Including no-code and low-code tooling as well as an access to the Fluxnova sandbox to provide developers Agentic orchestration capabilities. Actively encouraging Super Teams built across organizational lines. These are not small design decisions. They meant that domain expertise and technical skill could show up in the same room, on the same team, without one having to defer to the other. Some of the most interesting thinking I encountered came directly from those combinations: especially with the integration points across key FINOS initiatives like Fluxnova, the AI Governance Framework and CALM (architecture as code) to name a few was lovely to see teams leverage and implement in their core offerings. That kind of cross-pollination is what the industry talks about wanting and rarely actually creates the conditions for.

The open-source commitment through FINOS matters too. When code from this hackathon enters the FINOS ecosystem, it does not just benefit the team that wrote it. It becomes infrastructure that other firms, developers, and compliance functions can build on. The compounding return on that is significant, and it reflects a genuine belief that progress in financial services requires sharing across institutional boundaries, not hoarding within them.

Gratitude feels like too small a word for what DTCC and FINOS built here. They created conditions where the right problems got worked on seriously, by people with the right mix of expertise, with enough structure to focus effort and enough openness to let real ideas surface. That combination is harder to engineer than it looks

The talent working on cyber risk in financial services is not the constraint. I saw that clearly over five days. The ideas are there. The motivation is there. What accelerates everything is a platform that takes those ideas seriously from day one. This is that platform.

Get Involved: Build or Judge at the FINOS Fall 2026 Hackathon

The collaborative momentum continues this season at the FINOS Fall 2026 Hackathon. Whether you want to put code into production or bring your institutional perspective to our evaluation panels, there are two primary ways to participate:

• Build with the Community: Tackle real-world financial services challenges using open source standards and frameworks, including Fluxnova, CALM, and the AI Governance Framework. Continue the energy of OSFF New York by diving directly into project repositories and rolling up your sleeves.
• Join the Industry Judging Panel: Guide the next wave of open financial innovation. If you are an enterprise architect, security engineer, risk leader, or technology executive interested in evaluating working prototypes against production standards, our team wants to hear from you.

➡️ Register to Participate in the Fall 2026 Hackathon

➡️ Express Interest in Judging: Email us directly at help@finos.org

Author: Aditya Vikram Kashyap, Vice President, Firmwide Innovation, Morgan Stanley & FINOS Ambassador