Responsible AI
Effective 16 November 2026.
Intentric uses AI to turn a plain-language request into a proposed change and to answer questions about your deployments. This page says where, what the AI cannot do, and what happens to your data when it is used. The short version: the AI proposes, a person or your own rule decides, and everything is recorded.
Where AI Is Used
- Drafting a change. It turns your request into clarifying questions or a draft of what to build. The draft is checked against what the platform supports before you see it, and nothing is created until you proceed.
- Answering questions. It explains a deployment, its recent changes, and its verification and drift findings, to the member who asked.
- Writing runbooks. Through the API, it writes an operations runbook for a deployed set of infrastructure. A runbook is refused if it cites something that was not in its input or tells the reader to change a system directly.
In the console, a notice in the Intentric AI panel tells you that you are talking to an AI, and drafts and answers it produces are labelled as such.
What the AI Does Not Do
- It never holds the keys to your cloud. No AI component holds credentials for your cloud accounts. Only the step that applies an approved change gets a short-lived credential, for that job.
- It cannot approve or apply a change. A change is applied only after a recorded approval: a person’s, or your own auto-approval rule where you have turned one on, which never admits a change to something already live. The MCP server that lets your own agent drive the platform has no tool that approves or applies.
- It does not write the infrastructure code. The code is produced from the approved design by fixed templates, not generated by a model, so what you review is what runs.
- It cannot be talked out of the rules. Your limits are checked against what a change would do, not against the conversation that produced it. Rephrasing a request does not change the answer.
- It does not make decisions about people. The platform acts on infrastructure requests and is not built to decide anything about individuals.
How Your Data Is Handled
- Fixed instructions, your text as data. The instructions the AI runs under are fixed. What you write, and anything else that comes from you or your infrastructure, is passed in as material to read, never as instructions that can change those rules.
- Secrets are removed first. Before anything is sent to a model, text that looks like a credential, such as an access key, private key, password or token, is replaced with a placeholder. The check is pattern-based and not exhaustive, so please do not put credentials in a request.
- What is sent. Your request. For a question about a deployment, also that deployment’s resources and settings, its recent changes, its latest verification report and its recent drift findings. A question about your whole organization sends only the question.
- What is recorded. Each AI run is recorded: which provider and model answered, token use and estimated cost. The record does not contain the text of your request or the model’s reply. The platform also records which door a change came through: the console, the command-line tool, the MCP server or the API.
- Limits. Each run has a budget, and the service rate-limits requests.
Model Providers and Training
The hosted service reaches AI models through OpenAI, using its API. The provider is listed on the Sub-processors page.
We do not train or fine-tune models on your data. Under OpenAI’s API terms, data sent to the API is not used to train or improve OpenAI’s models, and OpenAI keeps abuse-monitoring logs for up to 30 days, unless longer retention is required by law.
Intentric monitors AI regulation, including the EU AI Act, and will update this page as obligations apply to the service.
Questioning or Reporting an AI Output
AI output can be incomplete or wrong, and it can differ between similar requests. Review every proposed change before you approve it; you can reject any of them. The Terms of Service set out your responsibilities for that review.
If an output looks wrong, harmful or surprising, or you think the AI was made to behave outside its role, tell us at privacy@getintentric.com, with the plan or request concerned. If you think you have found a way to get past an approval or a limit, report it through responsible disclosure instead.