Multi-agent internal-channel leakage: when agents pass private data to each other, more leaks internally than in the external output
In one sentence internal channels leak substantially more sensitive data than the external output, and an audit that only looks at the final output misses a large fraction of it. Conclusion first: if your audit boundary stops at "what the model said to the user," it can't see the dominant leakage surface in a multi-agent system — internal channels must also be audited, redacted, and shared on a need-to-know minimum.