Agent autonomously retries a blocked payment

What happens when an agent retries a blocked payment through another route, the evasion risk, and the control.

TL;DR

TL;DR: Retrying a blocked payment through a different address or rail is sanctions evasion under strict liability. A BLOCK must stop the flow, not trigger a reroute.

What happens

An agent's payment is flagged and blocked, but the agent is configured to retry through an alternative address, mixer, or rail until it succeeds. The retry turns a single blocked transaction into an attempt to evade the block itself. Each retry is a separate transaction, and each one inherits the same exposure.

The risk

OFAC enforces under strict liability, so the operator cannot claim the agent meant well. Civil penalties start at $356,000 per violation, and the 50 Percent Rule blocks entities 50% or more owned by a blocked person. Routing around a block, even once, is treated as evasion rather than a benign retry, and it multiplies the number of violations on record. A retry policy that ignores a block is an evasion policy, not an error handler.

The control

Make BLOCK terminal, not a signal to retry. agentmail returns BLOCK for a flagged counterparty in one call under 100 ms, and the SCREEN, SCORE, STOP, STAMP protocol halts the flow at the STOP gate so no retry path exists. dispute_open opens a documented review when a match is unclear, and the operator, not the agent, decides whether any alternative route is lawful. The STOP gate is what keeps an autonomous retry from becoming an autonomous violation.