Transparency

How DueFrame's AI works — and where it stops.

What generates the output

Reviews, answers, drafts and research briefs are generated by large language models from OpenAI, called over OpenAI's API at the moment you request an action. DueFrame pins the model versions it uses so behaviour cannot change silently between sessions, and records which model served each request. OpenAI does not train its models on API content. Your document text is processed transiently for the request and is not stored by DueFrame — the full retention picture is on the Security & privacy page.

What the AI is for

Reading the agreement in front of you and helping you act on it: flagging risks and gaps, proposing amendment wording, answering questions grounded in the document, summarising incoming redlines, checking documents against your playbooks, and researching legal points with sources. Quoted contract wording is always kept in its original language. Your saved personalisation and company profiles shape how answers read and whose side they take — they never loosen accuracy.

What it cannot be trusted to do alone

A language model can misread a clause, miss an issue, or state an inference with more confidence than it deserves. That is why every DueFrame output is drafted assistance, not a conclusion:

A human decides, always

No AI output is ever applied to your document automatically. Amendments land only when you click Apply, and they arrive as visible Microsoft Word tracked changes you can accept or reject. Comments and reminders are created only after your confirmation. When the AI fails, DueFrame shows the failure and offers a retry — it never silently substitutes something else and presents it as an AI result.

Where DueFrame sits under the EU AI Act

Under Regulation (EU) 2024/1689, DueFrame is the provider of an AI system built on a general-purpose AI model; OpenAI is the provider of the underlying model, and customer organisations use DueFrame as deployers. DueFrame's intended purpose is contract work by legal professionals in private practice and in-house. It is not intended for use by courts, tribunals or other judicial authorities, and is not marketed to them — it therefore does not fall within the Act's high-risk categories. The obligations that do apply at its tier are transparency about AI involvement and AI literacy, which this page and the disclosures across the product exist to meet. There is no certification scheme for this tier, so DueFrame does not claim a certification — just the practices described here.

Questions

If your organisation needs this in a procurement or vendor-review format, contact us and we will walk you through it.

Start free trial5 days free, no card
Get started