#multiagent-systems

Anthropic shows why agent swarms need new safety checks

Experiments with Claude agent teams found useful vulnerability discovery, but also coordination failures that can turn small model quirks into system-level risk.