Describe the laundering typology without being told laundering is illegal.

You cannot detect a pattern you are not allowed to describe. An analyst writing a rule for trade-based laundering needs the mechanics spelled out, which is the exact sentence a hosted model refuses.

Why the mainstream APIs fail here

Ask for a step-by-step description of how a sanctions evasion scheme is structured and the model treats the question as instructions rather than analysis. The same model will explain it happily from a news article, which tells you the refusal is about phrasing rather than risk. Meanwhile the analyst still has to file a narrative that explains exactly how the scheme worked.

How MagmaRouter is used for this

MagmaRouter answers the mechanics question directly, so typology documents, rule logic and SAR narratives get written in one pass. Per-key caps keep a batch of alert triage inside a budget. Prompts are not retained, which is the first question your own compliance team will ask about putting customer data near a model.

A working loop

  1. Cap a key to the triage budget for the period.
  2. Draft the typology and the detection logic with the mechanics stated plainly.
  3. Keep the receipt as the record that the alert text was not retained upstream.

Prompts are not stored. A signed receipt of the hash comes back on every call, so you can prove later what MagmaRouter logged. Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.

Related