Security
Your infrastructure stays yours. Creduna is designed so that the analysis works on metadata alone, through access that you create, can inspect line by line, and can remove in one step.
How access works
You create a cross-account IAM role in your own AWS account by launching a CloudFormation template that we publish. Creduna assumes that role to read data. We never ask for passwords or access keys.
The role trusts only the Creduna AWS account and requires an ExternalId that is unique to you. The ExternalId prevents another party from tricking Creduna into using your role, which is the standard protection against the confused deputy problem.
What the role can read
- AWS Cost Explorer: cost and usage by service and usage type.
- Cost and Usage Report metadata: which reports are defined.
- Amazon CloudWatch usage metrics: Bedrock invocation counts, input and output tokens, cache reads and writes.
- Amazon Bedrock model metadata: which models and inference profiles exist in the account.
- AWS Organizations: the list of accounts, to attribute cost.
What the role cannot do
- Read prompt content. It is not required for the analysis.
- Read completion content.
- Invoke your models or generate any cost.
- Modify infrastructure. The policy contains no write permissions.
- Read data from your databases, object storage or application logs.
The full policy is short. You can read the template before launching it at /creduna-readonly.yaml.

Revoking access
Delete the CloudFormation stack, or the role itself. Access ends immediately and nothing remains in your account.
Optional tools that use content
Two tools work on request content and are used only when you choose to provide it: the prompt cache analysis, which compares requests you paste, and the model comparison test, which runs a test set you supply. Pasted content is processed in memory and not stored. Neither tool is needed for the core analysis.
How Claude is used
Creduna sends Claude the computed usage profile: model names, token volumes, cache utilization, request counts and costs. It asks Claude to prioritize and explain. For application drafts, Claude receives the company profile you entered in the form.
For the Creduna Risk & Readiness Score, Creduna first checks facts in code: your answers, the program rules and one request to the home page of the website you entered, to see that it loads and what it says about the company. Claude then receives your answers without your name, email, country or AWS account ID, together with those verified signals and a short excerpt of the home page, and judges how clearly the application is explained and evidenced. Claude cannot change an eligibility status, and country is never used as a factor.
Creduna’s own Claude features run on the Anthropic API. Creduna also analyses Claude workloads that customers run through Amazon Bedrock. Anthropic and Amazon Web Services are listed as sub-processors in the Data Processing Agreement.
Creduna analyzes usage metadata such as model, token volume, cache utilization and service tier. Prompt and completion content is not required for AI FinOps analysis.
Connecting Anthropic
To analyse first-party Claude API usage, Creduna reads your organization’s usage report with an Anthropic Admin API key that you provide. This key is separate from the key Creduna uses for its own Claude features.
- The Admin key is used only to read the usage and cost reports. Those reports contain token counts and costs, not prompts or completions.
- The key is held in server memory for the duration of the analysis session and for at most one hour. It is not written to a database or a log, not returned to the browser, and not sent to Claude.
- You can revoke the key in the Claude Console at any time.
Our own systems
- Connections to the website are encrypted.
- The database is reachable only from the application’s internal network.
- The internal panel requires authentication.
- The website sets no analytics or advertising cookies.
Creduna does not hold security certifications at this stage. If your process requires a questionnaire, we will answer it.
Reporting a vulnerability
If you believe you have found a security problem, tell us before disclosing it publicly. Write to contact@creduna.co.