An AI that watches every keystroke, click, and file transfer is now worth $100M. INT’s new funding round, announced on July 20, 2026, marks a turning point in how enterprises approach security - not by chasing breaches, but by preventing them before they happen.
According to Brandon Dixon, CTO of INT, the current state of cybersecurity is broken by design. "We’ve accepted forensic analysis as the norm," he said on This Week in Startups. "But by then, the damage is done." INT flips the script by embedding lightweight reasoning models directly on employee devices. These models assess behavior in real time, stopping risky actions - like copying sensitive data into an unapproved AI tool - before they execute.
The system runs locally on CPUs, avoiding latency and privacy concerns from cloud-based monitoring. It relies on a dynamic "world model" of corporate policy, applying rules not as static documents but as live logic to every user interaction. The goal: eliminate the 99% of breaches caused by human error.
"We’re not just logging activity - we’re building a semantic substrate of how work actually gets done."
- Brandon Dixon, This Week in Startups
This deeper layer of observation has given rise to what the podcast calls "Work Obs" - a granular, behavior-level understanding of workflows. While such monitoring could feel invasive, INT argues it’s a productivity enabler. By analyzing who does what and how, companies can automate repetitive tasks and scale high-performing workflows across teams.
Dixon notes that oversight is adjustable: developers with high access might be under tighter monitoring, while legal teams can opt out. Hosting data on-premises further reduces privacy friction. The message is clear: you can’t optimize what you can’t see.
Meanwhile, Sume Labs is applying similar agentic logic to a different problem - AI video generation. David Im described the current state as a "generative slot machine," where users spin the wheel and hope for consistency. Sume’s API routes prompts across multiple models, stitching outputs into coherent, persistent narratives with stable characters and clean audio. It’s not just video generation - it’s production-grade content automation.
"We’re vibecoding a story, not just generating clips."
- David Im, This Week in Startups
