No, it is not safe to upload student work to ChatGPT or most cloud AI tools if that work contains personal information, unpublished research or anything covered by FERPA. The moment you paste text into ChatGPT, you send it to OpenAI's servers, where it may be stored, reviewed by staff for quality purposes and potentially used to improve future models.
This matters because student work often contains names, email addresses, institutional affiliations and sometimes sensitive information about learning difficulties, health conditions or family circumstances. Even if you think you have removed identifying details, metadata, writing style and contextual clues can make re-identification possible.
What happens to data you upload to ChatGPT
OpenAI states that conversations may be reviewed by human trainers and that data submitted through the free tier can be used for model training unless you opt out. The paid ChatGPT Plus and Team plans offer a control to prevent your data from training future models, but your text still travels to and sits on OpenAI's infrastructure. Enterprise plans add more controls, but even these do not eliminate the fundamental issue: you have sent student work outside your institution's control.
Other cloud AI services have similar arrangements. Google's Gemini, Anthropic's Claude and Microsoft's Copilot all process your input on remote servers. Some offer enterprise contracts with stronger privacy terms, but those typically cost thousands of pounds per year and require legal review. A teacher pasting homework into the free web interface has none of those protections.
FERPA and educational records
In the United States, the Family Educational Rights and Privacy Act prohibits schools from disclosing student education records without consent. Student essays, lab reports and exam answers are education records. Uploading them to a third-party AI service is disclosure. The fact that you remove the student's name before pasting does not always matter—if the work is still identifiable to that student, it remains protected.
FERPA allows schools to share data with vendors who perform institutional functions, but only under strict written agreements. A teacher using ChatGPT on their own initiative does not have such an agreement. The school has not vetted OpenAI, signed a data processing addendum or ensured that the vendor will not re-disclose the information. This creates liability.
UK and European institutions face similar constraints under GDPR. Student work that contains personal data—and most of it does—requires a lawful basis for processing and a data processing agreement with any third party. Pasting it into a web form does not meet that standard.
Even anonymised work carries risk
Teachers sometimes believe that removing names and student numbers makes upload safe. It does not. Anonymisation is harder than it looks. A student's essay about their grandmother's emigration from Kerala, their brother's diabetes or their village's water crisis is identifiable even without a name attached. Writing style itself can be a fingerprint—researchers have shown that machine learning models can attribute authorship with surprising accuracy.
If you upload 30 anonymised essays to check for AI-generated content or to generate feedback, and later that student's work appears in a legal dispute or disciplinary hearing, the question will be asked: who else has copies, and where did they go? The answer "I put it into ChatGPT" will not reflect well.
Terms of service and institutional policy
Many schools and universities have added clauses to their acceptable use policies that restrict or prohibit uploading student data to unapproved AI services. These policies exist because institutional leaders understand the legal risk. A teacher who ignores the policy and uploads student work anyway may face disciplinary action, especially if a data breach or privacy complaint follows.
Even where no explicit policy exists, employment contracts and codes of conduct generally require staff to protect confidential information. Student work is confidential. Sending it to a third party without authorisation breaches that duty.
What about detection tools
AI detection services such as Turnitin, GPTZero and others also require upload. Some are worse than ChatGPT from a privacy perspective, because they explicitly retain submitted text to build databases of known student work. Turnitin has been criticised for this practice—students objected that their intellectual property was being stored and used without meaningful consent.
Detection tools marketed to educators often have terms of service that grant the vendor a perpetual licence to use submitted content. Read the fine print before you upload a single essay. If the service is free, ask yourself how it makes money. The answer is often "your data".
Local AI eliminates the upload problem
The safest approach is not to upload student work at all. Offline AI tools that run on your own computer process everything locally. Nothing leaves your device. No account, no cloud, no third party with access to the data. You can analyse student code, generate feedback on essays or check for plagiarism without sending the work anywhere.
This approach also solves the consent problem. If a student submits work to you and you process it on your own machine, you are not disclosing it to anyone else. The data stays within the educational relationship. Schools and universities are far more comfortable with this model because it keeps them in control.
Local models are not as capable as the largest cloud models—GPT-4 and Claude 3.5 Sonnet remain stronger at complex reasoning and nuanced writing tasks. But for most teaching use cases—checking whether code runs, generating quiz questions, summarising readings—the gap has narrowed considerably. A local model that is 90% as good but keeps student data private is a better choice than a cloud model that is 100% as good but creates legal risk.
Alternatives for schools and institutions
If your institution wants to use AI at scale, the correct path is to negotiate an enterprise contract with a vendor, ideally one that processes data in your own jurisdiction and signs a proper data processing agreement. Microsoft and Google both offer education-specific AI products with stronger privacy terms than their consumer offerings. These cost money, but that cost reflects the legal and technical work required to handle student data responsibly.
Some institutions are deploying their own AI infrastructure using open models. This requires technical expertise and hardware investment, but it gives complete control over data. Schools and universities considering this path should budget for ongoing maintenance and model updates—self-hosted AI is not a one-time purchase.
Practical advice for teachers
If you are a teacher and want to use AI in your work, follow these steps. First, check your institution's policy on AI tools and data upload. Second, if no policy exists, ask your data protection officer or head of department before uploading any student work. Third, if you need to use AI for your own lesson planning or content creation, keep student work out of it—use synthetic examples or your own writing instead.
If you must analyse student work with AI, use a tool that does not require upload. Read the privacy terms carefully to confirm that processing happens on your device and that no data is sent to the vendor. Finally, document what you do. If a question arises later, you want to be able to show that you took reasonable precautions and followed institutional rules.
The convenience of pasting student work into ChatGPT is real, but so is the risk. Protecting student privacy is not optional, and ignorance of the rules is not a defence. Choose tools that keep data local, follow your institution's policies and ask permission before you upload anything that is not yours to share.