Researchers can apply for subsidized OpenAI API access through the Researcher Access Program
If you are interested in subsidized access to the OpenAI API for research, OpenAI asks you to describe your research use case on the Researcher Access Program application. Applying is an opportunity, not a guarantee: the policy says not all research will be prioritized for subsidy.
Supporting quotations
Researcher Access Program
If you are interested in the opportunity for subsidized access, please provide us with details about your research use case on the
Qualifications & exceptions
It is an "opportunity for subsidized access"; selection is not promised (see the review card).
Not stated here
The application form was linked but not supplied, so eligibility rules, the size of the subsidy and any terms attached to it are not known from this source.
Charges & Financial Obligations · Context
OpenAI lists research areas it considers especially important, but you can propose your own
OpenAI lists research directions it considers especially important for the program: alignment, fairness and representation, interdisciplinary research, interpretability and transparency, misuse potential, model exploration, and robustness. Each comes with example questions. You are free to craft your own direction instead.
Supporting quotations
There are a number of research directions we are excited to explore with the OpenAI API.
In particular, we consider the following to be especially important directions, though you are free to craft your own direction:
Alignment:
Fairness and representation:
Interdisciplinary research:
Interpretability and transparency:
Misuse potential:
Model exploration:
Robustness:
Qualifications & exceptions
The list is described as especially important directions, not a requirement: "you are free to craft your own direction".
Not stated here
The policy does not say whether proposals in the listed directions are more likely to be selected for subsidy.
Charges & Financial Obligations · Company practice
Applications take time to review, not all get a subsidy, and OpenAI only replies if you are selected
Because of a high volume of requests, OpenAI says reviewing Researcher Access Program applications takes time and not all research will be prioritized for subsidy. OpenAI will only be in touch if your application is selected, so you may not hear back at all if it is not.
Supporting quotations
Please note that due to a high volume of requests, it takes time for us to review these applications and not all research will be prioritized for subsidy.
We will only be in touch if your application is selected for subsidy.
Qualifications & exceptions
Contact happens only on selection; no rejection notice is described.
Not stated here
No review timeline or selection criteria are given.
This is the Sharing & publication policy, updated November 14, 2022
The saved page is OpenAI's Sharing & publication policy, marked "Updated: November 14, 2022". It sets rules for sharing AI-generated prompts and outputs online, publishing writing created partly with the OpenAI API, and publishing research about the API. This analysis covers the text as saved; it does not say whether OpenAI has since revised or retired this policy.
Supporting quotations
Updated: November 14, 2022
Sharing & publication policy
Qualifications & exceptions
Only the text in the saved copy of this page is analyzed; linked policies are not part of it.
Not stated here
Whether a later version replaced this one after the saved copy was captured is not known from this source.
The policy is written for the OpenAI API era (it names the Playground, the raw model and GPT-3); it does not name ChatGPT or say which current products it covers.
Content & IP Rights · Permission or license
You can generally post your prompts and outputs, livestream your use, or demo the products, if you follow the rules
Posting your own prompts or completions (the model's outputs) to social media is described as "generally permissible", and so is livestreaming your use or demonstrating OpenAI's products to groups of people. The permission comes with rules you are asked to follow: review each output yourself before sharing or while streaming, attribute it to your name or company, label it as AI-generated so no one could miss it, do not share content that violates the Content Policy or may offend others, and use good judgment with audience prompt requests. OpenAI says it set these rules to reduce the possible risks of AI-generated content.
Supporting quotations
Social media, livestreaming, and demonstrations
To mitigate the possible risks of AI-generated content, we have set the following policy on permitted sharing.
Posting your own prompts or completions to social media is generally permissible, as is livestreaming your usage or demonstrating our products to groups of people. Please adhere to the following:
Qualifications & exceptions
"Generally permissible" is not unconditional: the policy immediately says "Please adhere to the following" and lists the rules covered in the next cards.
The permission covers your own prompts or completions; it does not address sharing other people's prompts or outputs.
Not stated here
The policy does not say what happens if you share without following the rules.
The policy is written for the OpenAI API era (it names the Playground, the raw model and GPT-3); it does not name ChatGPT or say which current products it covers.
Content & IP Rights · Restriction or exception
When sharing outputs, review each one, put your name or company on it, and clearly mark it as AI-generated
If you share outputs on social media, in a livestream or in a demonstration, the policy asks you to manually review each generation before sharing it (or while streaming), attribute the content to your name or your company, and indicate that it is AI-generated in a way no user could reasonably miss or misunderstand.
Supporting quotations
Manually review each generation before sharing or while streaming.
Attribute the content to your name or your company.
Indicate that the content is AI-generated in a way no user could reasonably miss or misunderstand.
Qualifications & exceptions
These rules apply to the social media, livestreaming and demonstration setting; the publishing section has its own, similar attribution and disclosure conditions.
For livestreams the review may happen while streaming rather than before.
Not stated here
The policy does not give a required wording or format for the AI-generated label in this setting.
Don't share content that violates the Content Policy or that may offend others
When sharing outputs on social media, in livestreams or in demonstrations, you are told not to share content that violates OpenAI's Content Policy or that may offend others. The second part is broad: it reaches content that only "may" offend, not just content that breaks a written rule.
Supporting quotations
Do not share content that violates our [Content Policy](https://openai.com/policies/usage-policies/) or that may offend others.
Qualifications & exceptions
"May offend others" is not defined; the policy gives no test for what counts.
Not stated here
The Content Policy is linked (the link now points to OpenAI's usage policies page) but its text was not supplied, so what it forbids is not analyzed here.
If you take prompt requests from an audience, use good judgment and avoid prompts that might cause violations
If you take prompt requests from an audience (for example during a livestream or demonstration), the policy asks you to use good judgment and not to input prompts that might result in violations of the Content Policy. The duty is on you, the person entering the prompt, and covers prompts that "might" lead to violations, not only those that certainly would.
Supporting quotations
If taking audience requests for prompts, use good judgment; do not input prompts that might result in violations of our [Content Policy](https://openai.com/policies/usage-policies/).
Qualifications & exceptions
Applies when taking audience requests for prompts.
Not stated here
The Content Policy is linked (the link now points to OpenAI's usage policies page) but its text was not supplied, so what it forbids is not analyzed here.
Content & IP Rights · User right or choice
You may tell OpenAI about a particular completion by email or through the Playground reporting tools
If you want to make sure the OpenAI team is aware of a particular completion (output), you may email OpenAI or use the reporting tools in the Playground. This is optional.
Supporting quotations
If you would like to ensure the OpenAI team is aware of a particular completion, you may email us or use the reporting tools within Playground.
Qualifications & exceptions
Optional: "you may"; nothing requires you to report.
Not stated here
The source does not give the email address or describe the Playground reporting tools.
It does not say what OpenAI does with a reported completion.
Content & IP Rights · Company practice
OpenAI says you are using the raw model, which does not filter out biased or negative responses
The policy reminds you that you are interacting with the raw model, and that OpenAI does not filter out biased or negative responses. In other words, outputs you share or demonstrate can include biased or negative responses. It points to OpenAI's free Moderation endpoint as something you can read about implementing.
Supporting quotations
Recall that you are interacting with the raw model, which means we do not filter out biased or negative responses.
(Also, you can read more about implementing our
Qualifications & exceptions
The statement is made in the context of the 2022 API and Playground; the source does not say it applies to every OpenAI product.
Not stated here
The linked Moderation endpoint documentation was not supplied.
Whether this description still matches how OpenAI's current models are served is not stated in this source.
Content & IP Rights · Permission or license
You may publish your own written work created partly with the OpenAI API, under conditions
Creators who want to publish their own first-party written content (for example a book or a collection of short stories) created in part with the OpenAI API are permitted to do so, but only under the listed conditions: the work is attributed to your name or company, the AI's role is clearly disclosed, and its topics do not violate the Content Policy or Terms of Use. The policy also states that a human must take ultimate responsibility for what is published.
Supporting quotations
Content co-authored with the OpenAI API
Creators who wish to publish their first-party written content (e.g., a book, compendium of short stories) created in part with the OpenAI API are permitted to do so under the following conditions:
Qualifications & exceptions
The permission is conditional: "under the following conditions".
It covers first-party written content created in part with the OpenAI API; the source does not extend it to other media or to content made wholly by AI.
Not stated here
The policy is written for the OpenAI API era (it names the Playground, the raw model and GPT-3); it does not name ChatGPT or say which current products it covers.
The Terms of Use and Content Policy are linked but not supplied here.
Content & IP Rights · Restriction or exception
Published work must carry your name and clearly disclose the AI's role, e.g. in a foreword
To publish co-authored writing, the content must be attributed to your name or company, and the role of AI in creating it must be disclosed so clearly that no reader could possibly miss it and a typical reader would find it easy enough to understand. As an example, the policy says you must explain the relative roles of drafting, editing and so on in a foreword, introduction or similar place. You should not present the work as written wholly by a human, or wholly by an AI.
Supporting quotations
The published content is attributed to your name or company.
The role of AI in formulating the content is clearly disclosed in a way that no reader could possibly miss, and that a typical reader would find sufficiently easy to understand.
For instance, one must detail in a Foreword or Introduction (or some place similar) the relative roles of drafting, editing, etc.
People should not represent API-generated content as being wholly generated by a human or wholly generated by an AI
Qualifications & exceptions
The foreword or introduction is given as an example ("or some place similar"), not the only allowed location.
Both extremes are ruled out: neither "wholly generated by a human" nor "wholly generated by an AI".
Legal Rights & Agreement Rules · Restriction or exception
A human must take ultimate responsibility for content published with API help
For published content created partly with the API, the policy states that it is a human who must take ultimate responsibility for the content being published. You cannot shift that responsibility to the AI.
Supporting quotations
it is a human who must take ultimate responsibility for the content being published
Qualifications & exceptions
Stated for publishing co-authored written content.
Not stated here
The policy does not define what "ultimate responsibility" involves legally or practically.
Published topics must not violate the Content Policy or Terms of Use, such as adult content, spam or hate
A condition for publishing co-authored work is that its topics do not violate OpenAI's Content Policy or Terms of Use. The policy gives examples: the topics must not relate to adult content, spam, hateful content, content that incites violence, or other uses that may cause social harm.
Supporting quotations
Topics of the content do not violate OpenAI’s [Content Policy](https://openai.com/policies/usage-policies/) or [Terms of Use](https://openai.com/policies/terms-of-use/), e.g., are not related to adult content, spam, hateful content, content that incites violence, or other uses that may cause social harm.
Qualifications & exceptions
The listed topics are examples ("e.g."), not a complete list.
"Other uses that may cause social harm" is broad and not defined.
Not stated here
The Content Policy is linked (the link now points to OpenAI's usage policies page) but its text was not supplied, so what it forbids is not analyzed here.
The Terms of Use are linked but not supplied in this source.
OpenAI asks that you refrain from sharing outputs that may offend others
In the publishing section, OpenAI also "kindly" asks that you refrain from sharing outputs that may offend others. This is worded as a request rather than one of the listed publishing conditions, though the social media section states a similar rule as a direct instruction.
Supporting quotations
We kindly ask that you refrain from sharing outputs that may offend others.
Qualifications & exceptions
Worded as a request ("We kindly ask"), unlike the conditions listed before it.
Not stated here
"May offend others" is not defined.
The source does not say whether ignoring this request affects the permission to publish.
Content & IP Rights · Procedure
OpenAI offers stock wording to describe how you used the AI, if it is accurate
The policy provides stock language you may use to describe your creative process: it says the author generated the text in part with GPT-3, then reviewed, edited and revised the draft to their own liking, and takes ultimate responsibility for the content. You may use it only if it accurately describes what you did.
Supporting quotations
Here is some stock language you may use to describe your creative process, provided it is accurate:
The author generated this text in part with GPT‑3, OpenAI’s large-scale language-generation model. Upon generating draft language, the author reviewed, edited, and revised the language to their own liking and takes ultimate responsibility for the content of this publication.
Qualifications & exceptions
Optional ("you may use") and only "provided it is accurate".
The stock text names GPT-3; it would need to match the model and process you actually used.
Content & IP Rights · Permission or license
OpenAI welcomes research publications about the OpenAI API
OpenAI says it welcomes research publications related to the OpenAI API, because it believes the broader world should be able to evaluate its research and products, especially to understand and improve potential weaknesses and safety or bias problems in its models.
Supporting quotations
Research
We believe it is important for the broader world to be able to evaluate our research and products, especially to understand and improve potential weaknesses and safety or bias problems in our models. Accordingly, we welcome research publications related to the OpenAI API.
Qualifications & exceptions
Worded as a welcome rather than a formal license, and limited to research publications related to the OpenAI API.
Not stated here
The section does not set conditions such as review before publication; the possible actions OpenAI mentions are covered in the next card.
Security & Safeguards · Company practice
OpenAI may highlight your research, or act to protect users when it concerns security or misuse
In some cases OpenAI may want to highlight your research internally and/or externally. In others, such as publications about security or misuse of the API, it may want to take appropriate actions to protect its users. Both are stated as things OpenAI may want to do, not as commitments.
Supporting quotations
In some cases, we may want to highlight your work internally and/or externally.
In others, such as publications that pertain to security or misuse of the API, we may want to take appropriate actions to protect our users.
Qualifications & exceptions
"May want to": neither highlighting nor protective action is promised.
Not stated here
The policy does not say what "appropriate actions" are, whether they could affect your publication or your access, or whether you would be asked before your work is highlighted.
Cross-references
qualifies (model-proposed): OpenAI welcomes research publications about the OpenAI API — qualifies: OpenAI's welcome of research publications is paired with the statement that for security or misuse publications it may want to take actions to protect users.
Security & Safeguards · Procedure
Report safety or security issues you find in research immediately through the vulnerability disclosure program
If you notice any safety or security issues with the API while doing research, OpenAI asks you to submit them immediately through its Coordinated Vulnerability Disclosure Program.
Supporting quotations
If you notice any safety or security issues with the API in the course of your research, we ask that you please submit these immediately through our [Coordinated Vulnerability Disclosure Program](https://openai.com/policies/coordinated-vulnerability-disclosure-policy/).
Qualifications & exceptions
Worded as a request ("we ask that you please"), applying to issues noticed in the course of your research.
Not stated here
The Coordinated Vulnerability Disclosure Program policy is linked but not supplied, so its terms (for example any disclosure timelines) are not analyzed here.