Learn AI /AI Foundations /Proprietary model
Proprietary model
- Intermediate
- AI Foundations
- Policy officers
- Managers & front offices
Also called: closed model, closed-weight model.
Back to the milestone you came from
In five seconds
What it is
An AI model whose owner uses licence terms or technical controls to restrict access to or reuse of key components, such as its weights or code.
In plain words
A proprietary model is controlled through restrictive access or licence terms. Important components or rights to modify and redistribute it remain with its owner.
An everyday example
At an international meeting, a company provides a translation booth with skilled staff inside. Delegates speak into a microphone and hear translated audio through headsets. The service works, but the company does not allow anyone to enter the booth, inspect the notes, copy the method or hire another team to run the same setup. If the company changes its price or rules, the meeting organisers must accept that or switch services.
A human translation team can explain parts of its method, adapt intentionally and be directly supervised in real time. A proprietary AI model is software, usually operating through learned numerical parameters rather than human judgement. Also, some proprietary models can be run locally under contract, while the sealed-booth story suggests one fixed service setup.
Remember thisA proprietary model can be usable without being inspectable or reusable. Check which components and rights remain under provider control, not merely whether the service produces an answer.
This term is contested
The boundary of "proprietary model" is not fully settled. Some people use it broadly for any commercially restricted AI system, while others distinguish more carefully between closed weights, closed source, restricted licences and hosted-only access.
Understand it
Understand it
Why it matters
For decision-makers, "proprietary" signals a governance choice, not just a technical one. It affects procurement, cost, bargaining power, audit access, security review, data handling, business continuity and exit options if a supplier changes terms or stops a service. It also affects accountability after harm or error. If important parts are closed, an organisation may have less ability to inspect how the system was built, test it independently, reproduce results or move the workload elsewhere. That does not make proprietary models automatically bad or unsafe, and open models are not automatically better. It means the trade-offs must be examined clearly before adoption.
Proprietary model: what the user gets and what stays with the provider
After viewing this, the reader should be able to say that a proprietary model may allow use of outputs without giving open access to weights, code, training details or redistribution rights.
01
What the user sends
The user provides a prompt, document, image or other input through an app, website or application programming interface, often called an API.
Draw a user icon on the left with arrows to three small input boxes labelled text, image and audio, feeding into a box labelled service interface.
02
What the provider’s system does
The provider runs the model and related software on systems it controls, or under tightly controlled terms if customers can deploy it.
Place a central processing box between columns, shaded to show provider control, with sublabels model runtime and safety filters.
03
What the user receives
The user gets an output such as a summary, translation, score or generated image, but not necessarily the internal model itself.
From the central box, draw arrows back to the left toward output cards labelled summary, translation and score.
04
What stays closed
Key parts may remain unavailable for inspection, copying or modification, including weights, source code, training data information and redistribution rights.
On the right, stack locked boxes labelled weights, code, training details, hosting control and licence restrictions, each connected to the central processing box.
A two-column comparison. The left column shows what a user can usually access: input box, API or app, and outputs. The right column shows what often remains under provider control: weights, code, training data details, hosting choices and licence terms. Read from top to bottom to see the boundary between use and control.
Hovering or tapping each locked box reveals a short note on what that restriction can limit, such as auditing, portability or independent hosting.
Describe this without the diagram
This comparison diagram explains that a proprietary model separates use from control. On the left, a user sends text, image or audio through an app or API and receives outputs such as summaries, translations or scores. In the centre, the provider runs the model and related safety systems. On the right, several locked boxes show parts that often remain under provider control: model weights, source code, training data details, hosting choices and licence rights. The main point is that users may access the service without gaining full access to the internal model or the right to copy and redistribute it.
With reduced motion: A static two-column layout shows accessible user-facing features on the left and provider-controlled internal components on the right, with labels instead of hover effects.
A proprietary model is an AI model whose owner restricts access, use, modification, hosting or redistribution through technical or legal controls. The owner may let users send prompts or data and receive outputs without providing the weights or code. Some technical information may be disclosed, while licence terms still restrict what users can do.
What goes in depends on the service. A user might enter text, an image or audio. The model may run on the owner’s infrastructure or in licensed software under the owner’s terms. What comes out is a prediction, classification, translation, summary, image or other result. The user may see the output without being able to inspect or modify the model in detail.
This term is related to, but not the same as, open-source AI or an open-weight model. Open-weight describes whether trained parameters are publicly released. Proprietary describes ownership and restrictions on access or reuse. A model can therefore be both open-weight and proprietary if its weights are released under restrictive terms.
For leaders and policy staff, the key question is not only performance. It is also governance. Who can audit the system? Who can host it? What can be checked after an error? What happens if the owner changes the terms, prices or safety features? Those practical limits are often the real meaning of "proprietary" in use.
How “Using a sealed translation booth” maps onto the real thing
| In the story | In the AI system |
|---|---|
| Delegates speaking into the microphone | Users sending prompts or other inputs to the model |
| Translated audio in the headsets | The model’s output |
| The sealed booth and hidden notes | Private weights, code, training details or system design |
| Company rules about access and copying | Licence terms and provider control |
A worked example
An evaluation officer must test whether a proposed proprietary model can suggest the office's retention categories for monitoring reports without giving the provider access to live records.
Starting from: A test set of synthetic document titles and metadata containing no personal or confidential information.
- The officer submits the synthetic records through the vendor's hosted service and receives suggested retention categories.
- The vendor provides service documentation and test results, but does not provide the trained weights, source code or complete details of the training data and system settings.
- The office can test outputs and review contractual controls, but cannot independently inspect every part of the model or deploy that exact model on its own infrastructure.
- The officer records these limitations in the assessment and keeps all retention decisions with authorised staff.
Result: The test may show that the service is useful, but it also identifies provider-controlled components that affect auditability, hosting, continuity and any later move to another service.
In a United Nations workplace
During a records retention review, an evaluation officer in a Resident Coordinator's Office is assessing a vendor's proposal to classify monitoring reports before the file inventory goes to the front office at close of business.
The licence permits use through the vendor's hosted service but not access to the model weights, source code or full training-data provenance. That limits independent inspection, self-hosting and reuse, and makes future access, audit evidence and migration dependent on the provider.
Before you try thisEvaluation records may contain personal, confidential, human resources or procurement-sensitive material. They must not be uploaded unless the service is approved for that material and the applicable safeguards are in place, because the provider may process or retain inputs outside the office's controls. Model suggestions must not determine retention or disposal without authorised human review.
Where you may hear it
- A vendor demo during a procurement unit's technical evaluation
- An information technology governance board review of hosting options
- An internal audit meeting on access, logs and supporting evidence
- A records retention review before the file inventory is cleared
Avoid confusion
Avoid confusion
Often confused with
Open-source AI
AI systems released under licence terms that let others inspect, use, modify and redistribute source code and other materials needed to run them.
Open-weight model
A model whose trained numerical parameters are publicly released so others can run, inspect or adapt it, even if the training code, data or licence are not fully open.
Foundation model
A machine-learning model pretrained at large scale on broad data can be adapted for many tasks without training a new model from scratch.
Side-by-side
Common misconceptions
Common misconception: A proprietary model is simply a model you have to pay for.
Correction: Price and openness are different questions. Some paid services share more technical access than others, and some free services still keep weights, code and training details closed. The core issue is control over inspection, modification, hosting and redistribution, not only whether there is a fee.
Common misconception: Proprietary means the model is better or safer than open alternatives.
Correction: Closed access does not by itself prove higher quality or stronger safety. A proprietary provider may invest heavily in testing and safeguards, but outside scrutiny can also be limited. Quality and safety need evidence from evaluation, governance and suitable use, not assumptions based on business model alone.
Common misconception: If an API is available, the model is open.
Correction: An API lets people use a service interface. It does not usually grant access to the underlying weights, source code, training data details or redistribution rights. You may be able to send inputs and receive outputs while the most important internal parts remain closed.
Quick check
Which statement best describes a proprietary model?
The defining feature is control, not price or where the model runs. Some proprietary models are offered through APIs, some through licensed deployments, but in both cases important technical parts or usage rights remain closed or restricted.
Go deeper
Go deeper
The simple version
A proprietary model is controlled through restrictive access or licence terms. Important components or rights to modify and redistribute it remain with its owner.
The technical version
In technical and legal practice, a proprietary model is one whose access and use are governed by restrictive terms that keep one or more core components under provider control. Those components may include the trained weights, source code, architecture details, training data provenance, system prompts, fine-tuning methods, deployment stack or rights to copy, modify, self-host and redistribute. Users may interact through an API, application or licensed package, but they often cannot fully inspect the artefact, reproduce the training pipeline, run independent variants or transfer the model freely across environments. The exact boundary varies, which is why people sometimes distinguish closed-weight, closed-source and proprietary as related but not identical labels.
Reviewed external resources
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This UNESCO recommendation gives a policy and governance frame for assessing AI systems and suits leaders and public-sector readers thinking about accountability, transparency and oversight.
Intermediatelink checked 2026-07-31
EncyclopaediaWikipedia search: Proprietary model (opens in a new tab)Wikipedia
This Wikipedia search results page helps a reader scan for different uses of the term and is best for someone starting broadly, but it is not a single vetted article on one definition.
search results, not one pagelink checked 2026-07-31
ArticleMachine Learning Glossary (opens in a new tab)Google
This Google glossary is useful for readers who want short industry-style definitions of machine learning terms and suits busy professionals needing quick orientation.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Proprietary model (opens in a new tab)YouTube search
This YouTube search results page helps visual learners find a range of video explanations, but it is a search page rather than one reviewed source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Proprietary model (opens in a new tab)Reddit search
This Reddit search results page shows how practitioners and users discuss the term informally and suits readers who want to see common questions and points of confusion, with careful fact-checking.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
A proprietary model can be usable without being inspectable or reusable. Check which components and rights remain under provider control, not merely whether the service produces an answer.