Part C · CS-CF-05
AI Use and Transparency Statement
1. Why this statement exists
Cornerstone is built around an artificial intelligence assistant. That is the point of the product, and it means you are entitled to know exactly what the assistant does, what it does with your information and where you should not rely on it.
This statement is written to be understood by a site engineer, not only by a lawyer. It forms part of our Terms of Service.
2. Where AI is used in the platform
- Ask Cornerstone answers questions about how to use the platform and about your project information. It is given your question and the project content relevant to it.
- Document intake and filing classifies uploaded files and files them into the correct folder or register. It is given the file and its text.
- Data extraction reads values out of contracts, quotes, tickets, licences and tender breakdowns into structured records. It is given the file and its text.
- Contract review produces clarification schedules and requirements registers from a contract. It is given the contract document.
- Quote vetting compares quotes against a scope and highlights differences. It is given the quotes and the scope.
- Correspondence drafting drafts RFIs, TQs, letters and minutes. It is given your instruction and the related project content.
- Report narrative drafts commentary for cost reports and progress reports. It is given the underlying cost and programme data.
- Prequalification scoring scores a submitted prequalification against criteria the organisation has set. It is given the submission and the criteria.
3. How it works, in plain terms
The assistant uses a large language model. A large language model predicts text. It does not understand your project, it does not check facts against a source of truth, and it does not know when it is wrong.
When you ask it something, we send it your question together with the relevant information from your project so it has something to work from. It returns a response. We show you that response.
The model does not act on your project by itself. It does not issue documents, send correspondence, approve costs, change a programme or make a decision. A person does that.
4. What happens to your information
We send content to a third party model provider. To produce output we send the relevant content, which may include your project data, to our model provider. The provider, the country of processing and the retention period are set out in our Sub-processor List.
We do not train models on your content. Neither we nor our model provider uses your content to train a generative AI model. This is a contractual commitment, not a setting.
We do not mix customer data. Content from one organisation is never used to generate a response for another organisation.
We log interactions. We record the prompt, the response, the feature used, the model used and the tokens consumed. We do this for support, for billing, for abuse investigation and for quality review. These logs are retained in accordance with our Privacy Policy and Data Retention and Deletion Schedule.
You control what goes in. The assistant only sees content that exists in your project or that you attach to a conversation, and only within the permissions of the person asking.
5. Our design commitments
- Human in the loop. No AI output is applied to your project without a person reviewing and accepting it. Extracted data goes to a review queue. Generated documents are drafts until issued.
- Traceability. Where the assistant relies on a document in your project, it identifies that document so you can check it.
- No automated decisions about people. We do not use AI to make a decision producing legal or similarly significant effects on an individual without human review.
- Scoped access. The assistant operates within the permissions of the user who invoked it. It cannot surface content that user could not otherwise see.
- Change control. A change of model, provider or material configuration is treated as a release, is tested before it ships and is recorded.
- Metering transparency. Your AI usage is visible to you in the platform so you can see what is being consumed and by whom.
- You can work without it. Every core workflow in the platform can be completed manually. The assistant accelerates work, it is not a gate on it.
6. Limits and your responsibility
AI output is information, not advice. It is not legal, contractual, engineering, safety, financial, industrial relations or professional advice.
The assistant can be confidently wrong. In a construction context, the ways that matters most are:
- misreading a clause, a rate, a quantity or a date from a document
- missing a qualification or exclusion in a quote
- producing a plausible but incorrect summary of a contractual obligation
- generating a cost or programme narrative that is inconsistent with the underlying data
- omitting something that is not in the material it was given
You must have a competent person review AI output before you rely on it. That means before you price from it, issue it, submit it under a contract, use it to support a claim, use it to make a safety decision, or use it to make a decision affecting a person's employment or engagement.
To the maximum extent permitted by law, we are not responsible for loss arising from reliance on AI output that has not been reviewed by a competent person. Nothing in this statement excludes rights you have under the Australian Consumer Law that cannot be excluded.
7. Accuracy, testing and feedback
We test AI features before release against representative project documents, and we re-test when a model or provider changes.
If the assistant gets something wrong, use the report option in the assistant or contact info@cornerstoneapp.com.au. We record and review these reports. Where a report reveals a systematic problem we will correct it and, where the problem may have affected other customers, we will notify them.
8. Governance
Responsibility for AI in the platform sits with the Director. Our internal AI Governance Policy sets out how features are assessed before release, how risk is recorded, how models are changed and how incidents are handled. It is mapped to the ten guardrails in the Australian Government's Voluntary AI Safety Standard.
We will update this statement as the platform and as Australian AI regulation develop.
9. Questions
Questions about how a feature works, about privacy and data handling, or about an output you believe is wrong or unfair all go to info@cornerstoneapp.com.au.