Welcome to AI for FIs, from Dixon Strategic Labs. Each week, this newsletter curates critical developments in agentic AI and explains why they matter for your credit union.
Last week, an AI coding agent was supposed to write code. Instead, it modified its own behavior guidelines, published a personal attack on the developer who rejected its work, and ran autonomously for nearly three days.
It wasn't the only one. That same week, an AI agent hijacked GPUs to mine cryptocurrency, and a group of agents forged admin credentials while drafting social media posts. Deloitte found that even the largest U.S. banks don't have consistent governance over their AI systems. The stories below dig into what went wrong, what's being built to fix it, and one CU-specific vendor betting $25M on an agentic future.
Let's dig in.
IEEE Spectrum · Mar 10, 2026
An AI agent modified its own behavior guidelines and launched a public attack on a developer who rejected its code. The agent acted autonomously for 59 hours before its creator shut it down.
Why it matters: AI agents that rewrite their own guardrails put credit unions on the hook for actions no human authorized.
PYMNTS.com · Mar 16, 2026
Financial institutions deploying agentic AI face pressure to manage machine identities, strengthen cybersecurity, and ground AI in trusted data. Context engineering, the practice of feeding AI the right rules and business data, is emerging as a key requirement.
Why it matters: Deloitte found uneven executive control over AI-critical functions at the 28 largest U.S. banks, a governance gap credit unions should address early.
TechCrunch · Mar 16, 2026
Fuse, an AI-native loan origination platform, raised a $25M Series A led by Footwork. The startup also launched a $5M "rescue fund" offering 50 credit unions free access until their legacy contracts expire. Fuse already has over 100 customers.
Why it matters: Fuse is targeting credit unions specifically, offering a funded path off legacy LOS platforms like nCino and MeridianLink.
📡 On Our Radar
OpenAI launched GPT-5.4 with native computer-use mode and Excel/Sheets finance plugins, using 47% fewer tokens on some tasks.
The MCP roadmap through 2026 prioritizes security, authentication, and reliability fixes for production AI agent deployments.
BCG and OpenAI published a joint report on AI agents for retail FIs, projecting 30% profitability gains by 2030.
Alibaba's experimental AI agent autonomously hijacked GPUs to mine crypto, bypassing firewalls without human instruction.
Partnership on AI researcher argues cooperative governance models should guide AI infrastructure decisions at the community level. (More on this below - When Friction is the Point.)
⚙️ Tools & Vendors
Ramp Agent Cards — Corporate payment cards purpose-built for AI agents, with built-in spending controls.
Hiro Finance — A tool giving AI agents read-only access to personal financial accounts and transactions via secure OAuth connection.
This week we’re featuring a read that connects directly to the governance thread running through this issue. If AI agents are acting on their own, how should the organizations deploying them make decisions about oversight?
One researcher argues credit unions already have the answer.
“Human Structures (San Francisco), an art installation made up of 62 interconnected figures created by Jonathan Borofsky.” Photo taken by Bogdona Rakova in 2025.
Researcher Tamara Kneese makes a case credit union leaders should hear: slowing down technology adoption can be a strategic advantage.
In a recent interview on AI researcher Bogdana Rakova's Speculative F(r)iction Substack, Kneese argues that the rush to deploy AI often skips the human elements that determine whether a technology works. Friction, the disagreements, negotiations, and messy realities of implementation, is where the value is discovered and created.
Kneese highlights how communities (and cooperatives) that own and shape their infrastructure feel more empowered to use it well. She contrasts this with Big Tech's approach: build at scale first, ask questions later.
Involve your teams and members early. Co-design beats post-deployment damage control. Credit unions’ cooperative model gives you a structural advantage that large banks and fintechs can't easily replicate.
Friction is the work.
If you're rethinking how your credit union governs AI tools (or whether to give Fuse a call), I'd love to hear what you're weighing. Drop me a note at [email protected].
Know someone who should be reading this? Send them the subscribe link.
