The discussion around the use of generative AI tools in education is ongoing. With more institutions looking to implement these tools in classrooms and lecture theatres we wanted to investigate their user agreements. We have focused on two key questions to answer for now – whether the tool has an age requirement and whether the provider of the tool collects user inputs for training their models. We were keen to see what the current differences might be between a variety of generative AI tools and so looked at a selection of text, image and audio services. Of course, this technology is constantly evolving and changing so it’s worth noting that the information in this blog is correct at the time of our investigation.
We have looked primarily at tools which can be accessed free of charge as our focus is on those which are the most accessible to students. Many of these services do offer paid subscriptions and licenses which may have different terms to their free versions, and we may explore these in a later post.
The table below shows a selection of generative AI tools and their terms regarding age restrictions and using user inputs for training. This is by no means a complete list, but it can help us explore the terms under which different providers are offering their services.
|Generative AI Tool/Provider||What is the age requirement?||Can user inputs be used to train AI models?||Training LLM Explanation|
|OpenAI Products – ChatGPT, DALL∙E 2||18 or 13 with parent/legal guardian’s permission||ChatGPT
Yes – unless you opt-out
DALL∙E 2 – unclear
|Users can opt out of training in ChatGPT settings.
User inputs and outputs may be used to train models. There is no mention about switching off training in DALL∙E 2 settings.
|Google Bard||13||Yes||User inputs and outputs may be used to train models. Data where possible will be anonymised. Google have also reminded users not to include information that may identify the user.|
|Microsoft 365 Copilot||18 or 13 with parent/legal guardian’s permission||No||User inputs and outputs are not used to train the models.|
|Anthropic Claude||18||No – unless you opt-in||User inputs and outputs are not used to train the models. However, prompts and conversations may be used to train where explicit permission has been given. Such as by submitting feedback, ideas, or suggested improvements and they take reasonable efforts to de-identify user data.|
|Microsoft Bing Chat||18 or 13 with parent or guardian’s permission||Yes
|Non-enterprise version user inputs and outputs may be used to train models.
|Quillbot||18 or 13 with parent/guardian’s consent.||Yes||User inputs and outputs may be used to train models. Data where possible will be anonymised.|
|GrammarlyGo||13 in the USA with parent/guardian’s consent.
16 outside the USA.
|Yes||User inputs and outputs may be used to train models.|
|Wordtune||18 or 13 with parent/guardian’s consent.||Yes||User inputs and outputs may be used to train models.|
|UpWord||18 or 13 with parent/guardian’s consent.||No||User inputs and outputs are not used to train the models.|
|Scholarcy||13||No||User inputs and outputs are not used to train the models.|
|Midjourney||13||Unclear||We are unsure, we have reached out for comment.|
|Adobe Firefly – Photoshop generative fill feature||18||No||User inputs and outputs are not used to train the models.|
|AIVA||No age requirement||Yes||User inputs and outputs may be used to train models.|
|GitHub Copilot||13||No||User inputs and outputs are not used to train the models.|
We have found that most of the services we looked at required users to be a minimum of 13 years old though several, including Anthrophic’s popular Claude chatbot, require users to be 18. This may be an important point to consider when recommending use of these tools at FE institutions or for anyone with younger students. We only found one service with no age requirement at all, AIVA, a tool for generating soundtrack music. As with many online tools there is often little or no requirement of users to actually verify their age or prove they have parental consent before gaining access to these tools. Student users therefore might easily be in breach of the age requirements, whether that is done accidentally or on purpose.
Providers too have quite different approaches to whether they use user submitted data for model training. Many providers we found were clear in their terms on whether they use inputted data for training but some offered less clarity or did not mention either way. In those cases we reached out directly to those providers and most returned a definitive answer. We are still awaiting clarification on some and will update this post when they respond.
Interestingly, of those that do use user data to train only a few stated that they anonymise that data where possible and these still advise users to refrain from inputting personal information into their systems in the first place. Even where providers are attempting to keep personal data out of their training sets, there are no guarantees.
Overall, this investigation has demonstrated the wide variety in the terms and conditions around the generative AI tools available. As many users, both staff and students, will be engaging with AI tools independently of their institution and outside of a typical institutional/enterprise license agreement it is important to encourage all users to be aware of how data they input into these services might be used and to exercise caution.
Find out more by visiting our National centre for AI page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
For regular updates from the NCAI sign up to our mailing list.
Get in touch with the team directly at NCAI@jisc.ac.uk