Researchers simulate 34,000-agent defense swarm

Researchers ran a simulated defensive AI agent swarm consisting of 34,000 leaderless agents tasked with finding AI-accelerated attacks across three days of enterprise logs. The flat organization successfully found 12 of the 46 seeded attack procedures with an average detection time of 36 minutes, though researchers noted that a leader-worker structure performed better by reducing duplicated effort. Open article →