New AI Maps Overlapping Groups — and Sniffs Out Cyber Intruders

Researchers have unveiled a new AI system called DISCO that figures out which people, devices, or accounts belong together in a network — and crucially, it accepts that one thing can belong to several groups at once. Think of how you might be in your family group chat, your work team, and your weekend soccer club all at the same time. Older AI tools tend to force everything into a single box, which blurs the lines between groups. DISCO instead lets one node sit in multiple camps, giving a truer picture of how real networks work.

The clever part is how it looks around. Most AI models only pay attention to a node's immediate neighbours — the people directly next to it. That's like only knowing your desk mate at a huge office. DISCO adds a "spreading influence" step, borrowed from how rumours or diseases travel, so the AI can spot structurally important connections several hops away. It then focuses its attention on those links, which helps it detect group boundaries that would otherwise get smudged.

To show this isn't just theory, the team ran a cybersecurity demonstration. They took snapshots of a communication network over time and tracked how group memberships changed. When a device's group suddenly shifted in an odd way, that became a signal of possible trouble — and the method could point to which device was involved. The idea is appealing because the alert comes with a reason, not just a red light. Security teams could investigate a specific machine rather than guessing.

The honest caveat: this is an academic paper, tested mainly against standard benchmarks and a proof-of-concept security scenario. It hasn't been battle-tested on messy, real corporate networks at scale, and it isn't a product you can buy. Still, it points toward AI that understands groups better and explains itself more clearly — useful groundwork for fraud detection, social analysis, and network defence.