Multi-agent risks4

Claims and behaviors spread through peer networks sideways, along informal ties — diffusion research finds weak ties and small-world clustering carry information and practices across a network far faster than formal reporting lines.

Sources: granovetter1973, watts1998

Appears on: /pan-lab, /practice/peer-edge-governance, /practice/structured-dissent

A single automated rule set applied uniformly and without human review produced tens of thousands of correlated wrongful fraud determinations in the documented Michigan MiDAS case — one flaw repeating at caseload scale rather than averaging out.

Sources: michiganag2022, ieeespectruma

Appears on: /pan-lab, /practice/cross-model-checking, /practice/reconcile-copied-records, /practice/peer-edge-governance

Model behavior — including misaligned behavior — can propagate through model-to-model channels: research shows narrow in-context examples and inter-model interaction can induce broadly misaligned behavior in the receiving model.

Sources: afonin2026, panpatil2025, betley2026

Appears on: /pan-lab, /practice/cross-model-checking

Emerging agentic-AI governance frameworks treat inter-agent interaction as a first-class assurance surface, requiring explicit oversight of agent-to-agent couplings rather than per-model evaluation alone.

Sources: khan2025, hammond2025

Appears on: /pan-lab, /practice/peer-edge-governance