What is sensitive in a contract?
- Party identities — company and personal names, registration numbers, addresses, signatories and contact people.
- Commercial terms — amounts, pricing mechanisms, discounts, milestones and payment terms; these are trade secrets in the fullest sense.
- The deal's existence — sometimes the most sensitive fact is that two specific organizations are negotiating at all.
- Metadata — tracked changes and comments in a Word file expose your side's negotiating positions.
What's the risk of pasting a contract into an AI tool?
Most business contracts include a confidentiality clause — or a separate NDA signed alongside them. Sending the contract to an external AI vendor is a disclosure to a third party, and under most confidentiality language that is a breach regardless of what the vendor actually does with the data. Vendor policies differ and change — check the current policy before every use — which is why contracts sit high on the list of what not to paste into ChatGPT.
When individuals are involved — employees, contractors, customers — a second layer applies: their details in the contract are personal data under GDPR and similar laws, and transferring them to a vendor requires a legal basis.
What's safe and what isn't — a quick reference
| Action | Safe? | Why |
|---|---|---|
| Pasting an identifiable contract into an external AI tool | No | Third-party disclosure — NDA breach and privacy-law exposure |
| Asking a general question about standard clause wording | Yes | No identifying details or trade secrets involved |
| Analyzing a contract after consistent anonymization | Yes | Parties, amounts and identifiers replaced by tokens — the legal substance remains |
How do you review a contract with AI safely — step by step?
- Scan the contract — party names, registration numbers, addresses, amounts, unique dates and contact people.
- Replace consistently — the first party is always COMPANY_001, its signatory PERSON_001; the same name always gets the same token.
- Clean the metadata — comments, tracked changes and document properties that expose the negotiation history.
- Run the analysis — summarize obligations, spot missing clauses, flag risks, compare drafts.
- Map back locally — "COMPANY_002's obligation in clause 7" is translated to the real name on your machine, without it ever leaving.
Is an anonymized contract still useful for analysis?
Yes — because legal analysis rests on structure, not identities. A one-sided indemnification clause is one-sided when COMPANY_001 signs it too; an unbalanced payment schedule looks unbalanced in tokens. Consistent replacement preserves the relationships between the parties throughout the document — the principle explained in depth in the document anonymization guide. Lawyers carry an additional layer of duties — privilege and professional confidentiality — covered in AI for lawyers.