Six Blind Spots of Behavioral Enforcement in Open Agent Ecosystems

Behavioral enforcement in AI agent systems produces six structural blind spots that compound as ecosystems become more open, multi-agent, and multi-workflow. This paper enumerates those blind spots and argues that the governance response must be attestation — not observability. Observability records what the monitored channels show. Attestation produces independently verifiable, tamper-evident evidence of what happened, who authorized it, and what the exposure window was — evidence that external parties can examine without trusting the operator. The distinction matters because the failure modes documented here cause enforcement to actively degrade governance quality: the more rules enforced, the less visible real behavior becomes, and the more confident operators feel about a system that is increasingly opaque. This is the sixth and most dangerous blind spot. The path forward is not stronger constraints. It is a substrate that produces receipts.