RatedWithAI

RatedWithAI

Accessibility scanner

AI CopyrightAugust 31, 2026

The RFP You Pasted In Was Somebody Else's Confidential Document

Proposal work has two directions of legal exposure and AI drafting touches both. Going in, the solicitation you fed the model was handed to you under a use restriction. Coming out, the response you are adding to your reusable library may have no author the law recognises.

Two Directions
Input is somebody's confidential document; output is your unprotectable asset
Human Authorship
Machine-generated text has no author, so the proposal library may be thin on protection
Certified Text
In public procurement, proposal statements can be certified representations

Direction One: What You Put Into the Model

A solicitation is not public material just because it arrived in your inbox. Commercial RFPs routinely carry a clause limiting use of the document to preparing a response, prohibiting disclosure to third parties, and sometimes requiring return or destruction after award. Institutional and enterprise buyers attach a full NDA before releasing requirements at all, because the requirements themselves reveal roadmap, architecture, and budget.

Pasting that text into a general-purpose assistant is a disclosure to the model provider. Whether it is a permitted disclosure turns on two things that are easy to check and easy to skip: whether the confidentiality clause carves out service providers or contractors bound to equivalent terms, and whether your actual arrangement with the AI provider is a business tier that excludes submitted content from training and carries the confidentiality commitments the carve-out assumes. A personal consumer account almost never satisfies the second condition.

Direction Two: What Comes Back Out

Proposal teams do not write bid responses once. They build a library — approach narratives, differentiator language, methodology sections, risk-mitigation boilerplate — and reuse it for years. That library is a genuine asset, and its value assumes the company can stop other people from copying it.

Copyright protects works of human authorship. Text generated by a model, accepted with light editing, contributes nothing the law recognises as authored, and only the human contributions — original passages, and the selection and arrangement of the whole — carry protection. A library assembled predominantly from lightly-edited model output is therefore much weaker against a departing proposal manager or a competitor who lifts your differentiator paragraph than the team assumes.

Where the Real Exposure Sits, Ranked

Fabricated Past Performance and Credentials

CRITICAL RISK

Model-invented contract values, client names, headcounts, certifications, or reference details that reach the buyer as factual statements — and in public procurement, as certified ones

Confidential Solicitation Text Sent to a Consumer AI Account

HIGH RISK

Requirements, budget ranges, and incumbent details disclosed outside any contractual confidentiality chain, in breach of the use restriction that came with the RFP

Undisclosed AI Use Against an Express Representation

HIGH RISK

Solicitations that now require an AI-use disclosure or a human-verification attestation, signed without anyone reading the reps and certs section

Another Bidder's Language Reproduced

MEDIUM RISK

Recognisable industry boilerplate or a competitor's published proposal language surfacing in output, particularly in a market with few players and heavily-templated responses

Unprotectable Proposal Library

MEDIUM RISK

Years of accumulated differentiator content that carries thin or no copyright, limiting what you can do when it walks out the door

Why Public-Sector Bids Are a Different Animal

In commercial sales, an overstated claim in a proposal is a contract problem — it becomes a representation, and possibly a warranty, in whatever agreement follows. In public procurement, the same sentence can sit behind a certification, and a knowingly false certification carries consequences well outside ordinary contract remedies. This is the single reason proposal teams should treat every model-supplied number, date, client name, and certification claim as unverified until a human has checked it against a source system, regardless of how confident the surrounding prose sounds.

Proposal-Team Controls Worth Putting In Place

Practical, and mostly free — the expensive part is discovering you needed them after award.

Read the confidentiality and use-restriction clause of the solicitation before any text goes into a modelRequired
Route all proposal drafting through a business-tier AI account with training on submitted content disabled and confidentiality terms in placeRequired
Check the representations and certifications section for an AI-use disclosure or human-verification attestation on every bidRequired
Mark every model-supplied fact — contract value, date, client, headcount, certification — as unverified until checked against a source systemRequired
Keep past-performance and reference sections out of the AI drafting workflow entirely; populate them from records onlyPolicy
Record which sections were human-authored so the copyright in the library is at least identifiable laterDocumentation
Treat the proposal library as a trade secret as well as a copyright asset — access controls and confidentiality obligations do not depend on human authorshipStrategy
Add an AI-use clause to subcontractor and proposal-consultant agreements so their drafting inherits the same restrictionsContract

Audit your AI product's compliance exposure

RatedWithAI helps tech teams understand their compliance posture across accessibility, privacy, and AI regulation requirements. Start with a free scan.

Scan Your Product for Free →

Frequently Asked Questions

The solicitation says nothing about AI. Are we free to use it?

Silence on AI is not permission to disclose. The operative clause is usually the use restriction on the document itself, which predates AI and is drafted broadly enough to cover sending the text to a third party. Read that clause rather than looking for the word 'AI' — its absence tells you the buyer has not thought about it, not that the buyer consented.

Can we register copyright in a proposal that was partly AI-drafted?

You can register the work while disclaiming the machine-generated portions and claiming the human contributions, including original passages and the selection and arrangement of the whole. What you cannot do is register the full document as if a human wrote all of it. For most proposal teams the more useful protection is trade-secret treatment of the library, which does not turn on authorship at all.

Our AI vendor's enterprise terms say they don't train on our data. Is that enough for the confidentiality clause?

It addresses the largest concern but not the whole clause. A typical restriction prohibits disclosure to third parties, full stop, sometimes with a carve-out for contractors bound to equivalent confidentiality terms. No-training terms plus a confidentiality commitment usually fit that carve-out; they do not help if the clause has no carve-out at all, which is when you ask the buyer rather than assume.

What about using AI to analyse a competitor's published proposal?

Publicly released award documents are generally fair to read and analyse. The problem appears when the analysis output is close paraphrase or verbatim reuse of protected expression, and when the source was obtained under a restriction — for example a teaming partner's material or documents from a prior engagement. Analysing is not the risky step; regenerating is.

Related Guides