AI without the black box

Two things are true at once at Cloudpliant: your AI assistant can operate the entire screening process, and no language model ever decides a compliance outcome. AI-operable, not AI-decided.

What AI does not do here

Not a single classification in a Cloudpliant report comes out of a language model. Risk is determined by a transparent rule matrix with legal bases, and every outcome is reproducible from the recorded inputs. An auditor or regulator can recompute any decision. The full rules are on the methodology page.

That choice is deliberate, for three reasons. The Wwft requires you to justify every classification: when a supervisor asks, it must be traceable which rule applied and why. Language models are also probabilistic: the same input can produce a different outcome, and accountability cannot be built on that. And an outcome must be traceable to its source: every check records which register and which list version was consulted. Which registers those are is on the data sources page.

How we handle data along the way is in the legal documentation: security, privacy statement and the data processing agreement.

What your AI assistant can do

Cloudpliant runs an MCP server (Model Context Protocol), the open standard that lets AI assistants such as ChatGPT or Claude work with software safely. Ask your assistant to screen a new client and a frozen, shareable report is ready: sanctions lists, PEP registers and insolvency, with risk classification.

The same safeguards apply as in the rest of the product: personal API keys, team quotas, cost controls and a complete audit trail. Connection instructions are on the API page.