Client audit work on AI that keeps nothing

The question a raw US model key cannot answer

A consultant running ISO 27001 engagements through a raw Anthropic or OpenAI key is doing something no client contract says out loud: every policy draft, risk register entry, and piece of audit evidence pasted into that key leaves the EEA and, on the default routing, lands on US infrastructure, where the provider's standard commercial terms keep inputs and outputs for up to 30 days. Regional-processing arrangements exist at both providers, but they are negotiated per account; a standard self-serve key does not have one. That is a transfer you must evaluate under ISO 27001 control A.5.14 and justify under GDPR Chapter V, with a DPA, a transfer impact assessment, and a sub-processor list your client's security team can read. None of that is a reason to give up AI on audit work. It is a reason to stop routing client evidence through a key that makes the transfer the default.

What changes with the EU aliases

Endpoint

https://api.ismscopilot.com/v1/chat/completions

EU aliases

isms-fast-eu, isms-thinking-eu

Processing

Mistral's EU-bound inference host. No transfer of AI-model processing outside the EEA. Storage sits in the EU (Frankfurt).

Retention

Zero retention of request content. No training on your prompts or outputs. Usage metadata is kept for billing only.

Proof per response

x-isms-processing-region: eu response header

Price

$2.80 in / $8.80 out per million tokens, same as the global path. Prepaid credits from $20, per-key spend caps.

Compatibility

OpenAI-compatible chat completions. OpenAI-protocol scripts change the base URL and the model name; raw Anthropic Messages clients swap in an OpenAI-compatible client.

Your first EU-processed request

curl https://api.ismscopilot.com/v1/chat/completions \
  -H "Authorization: Bearer sk-isms-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "isms-thinking-eu",
    "messages": [{ "role": "user", "content": "Draft the A.5.14 information transfer policy for a 40-person SaaS company." }],
    "ismscopilot": { "frameworks": ["ISO_27001"] }
  }'

The response carries x-isms-processing-region: eu and names the framework modules injected into the answer.

The specialist part you were missing anyway

Swapping the region is the smaller win. The bigger one is what the endpoint knows before you send anything: curated framework modules, 100+ of them from ISO 27001 and SOC 2 to NIS 2, DORA, and the EU AI Act, injected at inference so you are not licensing and maintaining a standards corpus yourself. Pin up to eight modules per request, and every response discloses which ones it used. The catalog is public and dated, so a client can see what the model knows and how fresh it is. This is knowledge as the included part, not a prompt you hand-write and let go stale.

The math against a frontier key

List rates observed 2026-08-26: Claude Opus at $5.00 in / $25.00 out per million tokens on Anthropic's pricing page; isms-thinking-eu at $2.80 in / $8.80 out on ours. On compliance work that reads far more than it writes, that difference compounds across every engagement in the portfolio, and the EU path carries no premium over our global path. Two honest footnotes. Anthropic's prompt caching and batch pricing are real mitigations and belong in your math if your workload fits them. And we claim no quality superiority over Opus or any frontier model: the recorded head-to-head, Claude Code alone against Claude Code delegating to ISMS Copilot, was a pre-registered accuracy tie on the tested items. The case for switching is region, retention, included knowledge, and price. Not a quality trophy.

Prefer to stay inside Claude?

Many consultants do their real work inside Claude or ChatGPT, not in scripts. For that shape there is a second door: the Account MCP connector. Your agent calls create_conversation and send_message against your ISMS Copilot account, and with Advanced Data Protection enabled in the web app, all AI processing routes to Mistral with zero retention, on every plan including Plus at $20 a month. Usage rides your chat subscription, not a prepaid wallet. The ADP toggle lives in the web app, so set it once per account; workspaces created over the connector do not inherit it automatically.

Sources and product truth

• API sub-processor list

  The EU-alias processing claim, in writing

• API zero-data retention

  What is and is not stored on the Model API plane

• Credits and limits

  Prepaid credits, spend caps, welcome credit rules

• Model quality record

  Published evals with sample sizes and caveats

• Console pricing

  Live unit rates, source of truth; the console wins when prices change

• Anthropic pricing page

  Claude list rates, observed 2026-08-26

What this page is not claiming

EU processing is a routing choice you make per request, not a company-wide hosting claim: the global aliases (isms-fast, isms-thinking, isms-mini) process in the United States, and we disclose that the same way we disclose the EU path. The endpoint is text in, text out: no tool calling, no JSON mode, no multimodal input; streaming is supported. There is no bring-your-own-key option; the key is ours and the spend cap is per key. And the output is compliance guidance, not an audit opinion: the consultant signs the deliverable, exactly as before.

Frequently Asked Questions