Larkup LogoLarkup
Back to Home

Safety Center

Safety on Larkup

Larkup is built around providing secure AI infrastructure for educational institutions. Users on the platform are connected to a specific university, and if they are taking part in automated academic workflows, they should always have the contact details of their IT administrator or the relevant academic office.

In all cases, AI deployments are verified by the university or programme team before they are integrated. That reduces risk and gives the educational system a clear structure. Even so, no platform or verification process removes the need for human oversight, clear communication, and institutional accountability.

Most interactions with Larkup's AI are highly productive. Still, unexpected outputs can happen. Educational workflows involve diverse materials, contexts, and learning styles. What seems like a logical AI response in one context may require refinement in another. That is why it matters to review outputs carefully, communicate clearly with your students, and pay attention to system boundaries.

Deploying AI for the first time

In educational automation, the first deployment often happens when an institution integrates AI into a specific course or administrative task. That can be completely transformative and helpful, especially when dealing with large volumes of grading or content generation.

Even in that situation, it is worth keeping the integration simple and clear. Confirm the parameters and permissions in advance. Choose an easy starting point, like syllabus generation or automated quiz creation. Make sure both educators and IT staff know how to monitor the system.

If you are an educator, make sure you know the capabilities of your AI tools, the contact details of your IT support team, and how to verify AI-generated outputs. If you are an administrator, be clear about what support you can realistically provide and do not make promises the technology cannot keep.

A first AI deployment can be supportive and practical. It should not involve pressure, confusion, or loss of control over the curriculum.

Data privacy, oversight, and boundaries

A good AI integration depends on strict data privacy and oversight. That includes respecting intellectual property, student data privacy, academic integrity, and institutional guidelines.

Not every unexpected AI output is a serious safety issue. Sometimes models misinterpret prompts. Sometimes expectations are different. When that happens, it helps to refine the prompt or adjust the workflow rules clearly and calmly.

At the same time, you should not ignore system behavior that compromises data security, student privacy, or institutional standards.

You do not have to share sensitive student information with public models, give third-party apps access to your private documents, or accept automated decisions without review. You are allowed to require human-in-the-loop verification. You are allowed to step back and audit the system. You are allowed to restrict access.

Automating education does not mean having no boundaries. Adopting new technology does not mean having to accept workflows that feel wrong.

If an issue arises

Take it seriously.

If the issue is a simple hallucination or incorrect output, try to make your prompt constraints clearer. If the system respects it, the situation may be resolved. If it does not, or if the output is inappropriate from the start, do not continue using that specific workflow without adjusting the underlying configuration.

If you experience system behavior that is biased, discriminatory, insecure, or otherwise inappropriate, contact your university IT department or AI coordinator. You can also contact Larkup support at support@larkup.com.

This includes situations such as repeated hallucinations, boundary-crossing data requests, generation of inappropriate content, or anything else that makes you feel the system is unsafe or undermines academic integrity.

If you are part of a university pilot program, your AI coordinator should be one of your first points of contact. They are there to deal with issues connected to the deployment and to help respond when something has gone wrong.

Reporting critical vulnerabilities

If a security boundary has been crossed, report it. Do not assume it is too small to mention.

Please report concerns to:

If possible, include what happened, when it happened, which models or workflows were involved, and any logs or screenshots that may help explain the situation.

If the matter involves a severe data breach or immediate security threat, do not wait for a response from the platform. Escalate the issue to your institution's cybersecurity response team immediately.

A final note

Larkup is designed to support secure, institution-linked AI integrations for educators and students. In most cases, these tools are helpful and straightforward. Still, safety also depends on how institutions implement and oversee them in practice.

Maintain oversight. Be clear with instructions. Be aware of data boundaries. Do not ignore system behavior that feels wrong.