ChatGPT IP Issues FAQ

ChatGPT is a generative AI technology. It can create a wide variety of output based on input received from the user. It was launched in November 2022 by OpenAI, an American artificial intelligence research organization. As of the date this FAQ was written, a "Free," "Plus," "Team," and "Enterprise" version of ChatGPT is available, with varying features in each of the different subscription models. The latest model of ChatGPT available as of the date this FAQ was written is GPT-4o, which OpenAI states can reason across audio, vision, and text in real time and is a step towards more natural human-computer interaction.

It is important to note that ChatGPT is not the only generative AI technology of its kind available, with Google’s Gemini, Microsoft’s Copilot, MetaAI, and Claude among some of the most popular alternatives. While many of the issues discussed in this FAQ may also broadly apply to the use of these alternatives, they may not apply in the exact same way due to differences in operation of the alternatives, as well as differing legal terms related to their use.

ChatGPT is a fine-tuned large language model that was originally trained to produce text output but is now also capable of producing output in a variety of other formats. At a basic level, ChatGPT attempts to understand input received by the user and then outputs a string of words and phrases that it predicts will be the best at responding to the input, based on the data it was trained on. According to OpenAI, ChatGPT was trained and developed using information that is publicly available on the internet, information that is licensed from third parties, and information that users or human trainers provide. With respect to information that is publicly available on the internet, this refers only to information that is freely and openly available, and does not include, for example, information behind paywalls. Presumably, most other sources of information are therefore fair game for the training of ChatGPT.

Furthermore, ChatGPT was specifically optimized for human-like dialogue by using the method of Reinforcement Learning with Human Feedback, a method that uses human demonstrations and preference comparisons to guide the model towards desired behavior. In this case, because the models were trained on vast amounts of data from the internet written by humans, including conversations, ChatGPT’s desired behavior and output are human-like responses.

From its inception, ChatGPT could be used in a variety of different ways depending on the context and user’s needs. Below is a non-exhaustive list of some of the ways ChatGPT has been reported to be used:

In brief, ChatGPT can be used in a vast number of different ways. In addition, with every new iteration of the model, the ways in which ChatGPT can be used appear to be increasing and becoming increasingly more diverse and creative.

Despite the numerous ways ChatGPT can be used to assist users, it has several limitations that users should be aware of. First, ChatGPT is not connected to the internet, and the current version of ChatGPT available to the public was trained using sources up to 2021. Therefore, it may not be helpful to users seeking more recent information. Second, because of the variety of sources that are used to train the model, it is possible that the output it provides contains errors or biases. Third, ChatGPT frequently does not provide sources for the information contained in its output, and if asked to do so, sometimes provides inaccurate or incorrect sources. Other limitations of ChatGPT to note include its potential inaccuracy, lack of logical reasoning, inability to understand context, as well as the usage limits for the free version of the model.

While not exhaustive, these limitations highlight just some of the reasons why OpenAI has several disclaimers in their Terms of Use related to the ChatGPT service, including that:

In general, ChatGPT collects data from users in order to further train the model and improve its performance. ChatGPT collects data such as user input, chat history, and user preferences from the user. Conversations between ChatGPT and the users are reviewed both by AI trainers to improve their systems and are reviewed to ensure that the content complies with OpenAI policies and safety requirements. Users of ChatGPT can see a history of their conversations within the interface, and although specific prompts cannot be deleted from the history, user are able to delete their data following the process laid out on the service.

More recently, OpenAI has stressed that they do not actively seek out personal information to train their models, but that personal information is incidentally included at times. OpenAI stresses that any personal information will be used exclusively to train the model and will not be used to build profiles of people or attempt to contact or advertise to them. Additionally, OpenAI states that they take several steps to limit the use of personal information and comply with privacy laws. With respect to the former, OpenAI removes websites that aggregate large volumes of personal information, as well as trains their model to reject requests for personal information. With respect to the latter, training data is stated to be obtained and used lawfully. In addition, users in certain jurisdictions can object to the processing of their personal information by the model in OpenAI’s privacy portal.

ChatGPT falls under the umbrella of generative AI technologies, which is used to describe software that can produce output in various forms, such as text, images, and audio, based on input from a user. Generative AI technologies have become increasingly popular in recent years, and due to their accessibility and the quality of output they are able to potentially produce, they are frequently used in both private and commercial contexts to create, or assist in creating, a variety of different works.

Some of the key questions regarding the intersection of intellectual property law and the use of technologies such as ChatGPT, therefore, includes who owns the output that is created by ChatGPT, whether the output could be the subject of a copyright or trademark infringement claim, and to what extent ChatGPT can be used in the process of logo creation and branding of a business. This section of the FAQ will explore these questions.

It is currently unclear who owns the copyright to content generated by an AI model, such as the output from ChatGPT. Because the Canadian Copyright Act (the "Act") does not explicitly address generative AI creation, many copyright issues pertaining to AI-generated content remains unresolved. One of the key issues in this space is whether the user can be considered an author for the purposes of the Act, or if the AI program itself should be granted authorship. Without authorship, there is also no corresponding ownership for copyright purposes.

This issue of authorship and ownership arises because it is unclear whether the user, by providing the input, would satisfy the requirement that under Canadian copyright law, a work must be original. To be original, the creation of a work must be the result of an exercise of both skill and judgement. Due to the current lack of clarity, the issue of authorship may ultimately be settled based on the degree of human involvement in the process to create the output.

In the 2023 Consultation on Copyright In The Age of Generative Artificial Intelligence (the "Consultation") published by the Government of Canada, the topics of authorship and ownership of AI-generated works were discussed at some length. The Consultation highlights that although the Act does not define "author," jurisprudence suggests that authorship is attributed to a natural person who exercises skill and judgement in creating the work.

The government is specifically seeking the views of stakeholders on three approaches to authorship and ownership of AI-generated works:

These approaches are not exhaustive, with the Government specifically stating that the possibility of other approaches is not foreclosed. The question of copyright ownership and authorship of AI-generated works in Canada is still very much unresolved.

Under the “Content” section of the Terms of Use, OpenAI states that with respect to “Ownership of content, as between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.” Therefore, it appears that OpenAI does assign the rights to the AI-generated content to the user.

However, the language of “to the extent permitted by applicable law” adds uncertainty, especially given the lack of clarity in copyright law. Furthermore, it is possible that no rights are actually assigned if ownership is settled “as between you and OpenAI,” as stipulated by the Terms of Use.

Yes. As outlined in the Copyright Act, s. 27, “it is an infringement of copyright for any person to do, without the consent of the owner of the copyright, anything that by this Act only the owner of the copyright has the right to do.” This includes selling, distributing, exhibiting, or otherwise reproducing a copyrighted work.

Because of the data ChatGPT relies on to generate responses, its output may at times be similar enough or identical to existing copyrighted material to constitute infringement.

The 2023 Consultation highlights the difficulties in establishing such a claim in Canada:

While there is no Canadian precedent yet, users are advised to review and, where appropriate, revise AI output before using it.

In Canada, it is still unclear whether legal action can be taken against AI companies for using data without permission, as courts have not yet ruled on this issue.

In the U.S., however, several lawsuits have been filed:

While Canadian law differs, these cases may provide guidance. For now, OpenAI’s Terms of Use allow users to submit notices of claimed infringement, which may result in removal of content or termination of accounts for repeat infringers.

Yes, but caution is required. A new trademark cannot be confusingly similar to an existing one under the Trademarks Act. If a ChatGPT outputted name or logo is too close to an existing trademark, it may not be registrable, and using it could result in trademark infringement.

This could lead to costly disputes or re-branding. It is recommended that users carefully review AI-generated branding and, ideally, consult an intellectual property lawyer to assess risks before adoption.

As part of OpenAI’s commitment to safe and responsible AI, conversations between the user and ChatGPT are reviewed to improve the service and to ensure that the content complies with their policies and safety requirements. Furthermore, conversations may also be used to further train the models. Therefore, to some extent, confidentiality is lost as soon as a conversation with ChatGPT has started.

Although OpenAI provides options to the users to attempt to remedy this, such as opting out of training through their privacy portal, by giving the option to turn off training for their ChatGPT conversations, or by disabling chat history, these options come with their shortfalls. If a user chooses the option to turn off chat history, for example, OpenAI will still retain new conversations for 30 days and review them if needed to monitor for abuse, before permanently deleting them.

Therefore, it seems that no matter what privacy option is selected by the user, OpenAI will potentially have access to the conversation regardless. The implication of this is that any personal, confidential, or sensitive information input by a user may be accessed by OpenAI and potentially appear in future outputs to other users.

Users should therefore refrain from inputting any personal, confidential, or otherwise sensitive information into ChatGPT. While it would be prudent for all users to follow this suggestion, it may be especially pertinent for users who regularly handle private and confidential information, or who have personal or ethical obligations. For example, users who are under nondisclosure agreements, who hold confidential information, trade secrets, or other proprietary information, should be especially careful to ensure that they do not use ChatGPT in relation to this information in a way that would lead to a loss of confidence or a breach of professional and/or ethical obligations.

Each user should be aware of, and factor in, the limitations and potential privacy concerns of ChatGPT when deciding if it is safe to use ChatGPT in their specific context. Users should also be careful to use the technology in a way that minimizes risk as discussed throughout this FAQ.

On the service side, OpenAI takes a variety of security measures related to ChatGPT. These measures include audits, data encryption, and access to a security portal which highlights many more of their security measures, including risk-profiles, data security, app security, data privacy, endpoint security, policies, and so on. That said, OpenAI, like most other service providers, is susceptible to data breaches. In March of 2023, for example, OpenAI reported having a bug in their open-source library which allowed unauthorized users to see the beginning of someone else’s conversations, account details, and even the last four digits of credit cards. The possibility of a data/security breach is another factor users need to consider when determining whether ChatGPT is safe for their specific case.

On a system-wide level, although the Canadian government has not taken an explicit stance, it is important to note that in April 2023, the Office of the Privacy Commissioner of Canada launched a joint investigation into OpenAI’s ChatGPT with the provincial privacy authorities of Quebec, British Columbia, and Alberta. The investigation was launched in response to a complaint alleging the collection, use and disclosure of personal information without consent.

The privacy authorities will investigate, among other things, whether OpenAI:

As of the date this FAQ was written, the investigation is ongoing.

The potential risks surrounding the limitations, privacy, and confidentiality concerns of ChatGPT have been canvassed throughout this FAQ. It would be prudent for a business that is contemplating the incorporation of ChatGPT into their operation to assess the variety of privacy-related risks that exist within their workplace and use cases, and subsequently, consider in what ways ChatGPT should be restricted in the workplace, if not prohibited.