Image generated using Flux AI (Dev)

Generative AI models and applications are improving at a rapid pace. It is difficult to keep up with the capabilities and functionalities of these. The technology and its applications have advanced to a stage where any assessment that is not supervised and in-person could be fairly readily performed by generative AI.

This is not to say that unsupervised assessments are not worthwhile. That is, unsupervised assessments still play an important role in students' learning - but they cannot be used to assure program learning outcomes have been attained.

Here we outline some key examples of how assessments that are outside of Secure (Lane 1) assessments might be completed using AI. New AI tools are released almost daily and are actively advertised to students through social media and other channels.

Practice or application: In-class or out of class quizzes

Quizzes are an excellent way for students to practice or apply their knowledge and skills in order to get feedback on their development. However, there are browser plugins, local AI models, and even AI agents now available that can complete these assessments.

Browser plugins exist that can embed themselves into the quiz pages of Canvas (and other LMS), read the quiz questions, and answer on behalf of the student. The image at the right shows one such plugin, and you can see it in action here.

Cutting off the internet is also not protective as there are many AI models that can be run 'locally' without needing internet access, through tools such as Ollama. These tools are actually quite convenient when one cannot connect to the internet and/or wants to keep data secure.

Late 2024 saw the introduction of AI tools that could take control of a browser and, using AI vision, navigate web content and complete tasks - including assessments. Early 2025 saw the introduction of OpenAI's Operator AI agent, which has been shown to handle Canvas quizzes easily with very little user intervention (or thought).

Inquiry or investigation: Data analysis

Data analysis is an important skill in many disciplines, where students may gather their own data and then analyse it to present as part of a report. These can be very authentic assessments, and once were considered secure especially if students were using data not available elsewhere.

Since 2023, ChatGPT has had the ability to analyse uploaded data (either through plugins, or natively). A spreadsheet of data can be provided to the AI along with a bit of background about the data, and the AI can open and interpret the spreadsheet, perform analyses, visualise data, and suggest further investigations. The image to the right shows such an analysis, and you can see this in action here.

Inquiry or investigation: Case studies

Analysing real-world scenarios to identify problems, propose solutions, and justify decisions is a key application of critical thinking and professional readiness. Often these case studies draw from material that is provided within a unit of study, and may be unavailable elsewhere.

These were once considered 'AI safe' if the material was not accessible online to the public. However, many AI tools exist that can easily use information from uploaded documents to inform responses to users - including widely-used AI tools like ChatGPT, Gemini, and Claude. Many of these tools have increasing 'context window' sizes which means the amount of text they can absorb is in the order of hundreds of thousands to millions of words - much more than a student could.

Some AI tools exist primarily for the purposes of being able to take in many documents for summarisation and helping the user find interconnections between the documents, such as Google's NotebookLM. You can see this in action here, working on 'private' documents.

Inquiry or investigation: Research analysis

Being able to critically examine and interpret research sources and findings is a key part of all scholarly disciplines.

In the early days of ChatGPT, it was widely known that it would 'hallucinate' citations and sources - that is, it would confidently make up fake sources. However, since later in 2023, ChatGPT and some other AI tools have been able to draw on real scholarly sources. These AI tools can access the literature and compose convincing literature-informed pieces - you can see one in action here. Some of these tools are even officially sanctioned by publishers and institutions.

Towards the end of 2024, a number of AI companies released tools called 'Deep Research' or its variants, including Google, OpenAI, and Perplexity. These tools were early examples of 'AI agents' - AI that could be given a task (e.g. analyse the impact of the rise of social media on the economy) and be left alone for a few minutes to 'think', look up and read sources (including scholarly sources), consider an analysis, and compose an informed response complete with tens to hundreds of citations. It's fascinating to watch these deep research AI agents work - they have internal monologues, search the internet and other sources, activate their internal capabilities like coding and analysis engines, and more to produce a response to your task.

Production and creation: Portfolio or journal

Being able to demonstrate learning and growth through producing and curating samples of work, reflections, and other evidence plays a key part in laying out the process of learning. Early suggestions to 'AI proof' assessments included advice to include personal reflections because it was thought that AI could not produce these.

As shown in other assessment types, the samples of work themselves could be produced by AI with little to no contribution from the student.

Reflections are also very easy for AI to emulate. There are even tools available that specifically target reflective writing - see some in action here.

Production and creation: Performance or presentation

The creation and delivery of live or recorded performances or presentations can be an authentic task for many disciplines. Early advice to 'AI proof' assessments was to introduce a mandatory live presentation or performance, delivered online or in person.

However, generative AI tools can be easily used to generate material ahead of time for these outputs - including the script, slides, and more. See this in action here, using a few different generative AI tools. Students need only ask generative AI to write a script, and ask it to generate a slidedeck, and memorise this to perform in front of a class.

For multimedia artefacts like video and audio, a number of generative AI tools exist now that can produce video, clone voices, make video avatars of yourself, generate music, and more.

Production and creation: Written work

Developing and producing a structured, thoughtful piece of long form writing has been a mainstay for higher education for centuries. Intrinsically, long form writing has a critical role to play in helping students develop deep and critical thinking skills - after all, writing is thinking. However, from the early days of ChatGPT, it was clear that the security of written work was at significant threat from AI.

It is now widely known that countless generative AI tools, including the main platforms like ChatGPT, Gemini, and Claude, can produce high quality written work with the right set of prompts - which may include the assignment brief, the rubric, uploaded private sources, and more.

Production and creation: Dissertation or thesis

These written manuscripts are often the culmination of an extended and original research project. Their functions include to engage the student in deep critical analysis and integration of scholarship into their findings. However, as described above, a number of generative AI tools are now able to perform this work with potentially little contribution from students themselves.

Discussion: Debate, Contribution, or Conversation

These assessments enable students to share knowledge, evaluate and defend their own and others' ideas, and create shared outcomes. These are crucial learning objectives for all students.

However, generative AI is able to emulate these easily in asynchronous environments such as discussion board posts. A student only needs to copy and paste the discussion so far, plus the brief, into a general purpose AI tool and it will happily generate a response.

In synchronous (live) online environments, there are generative AI tools available that can transcribe the spoken conversation, and use AI to suggest a response for the student to offer verbally. See one in action here, also shown in the screenshot to the right where the transcript appears on the left of the window and the suggested response appears on the right.

Discussion: Evaluation

Developing evaluative judgement - the ability to judge quality - is a critical outcome of a higher education. Students should be able to evaluate the quality of their own and others' work and make informed judgements based on their knowledge and skills.

As suggested above, in online environments it is straightforward for students to copy and paste work into a generative AI tool, along with the evaluation criteria such as a rubric, and ask the AI tool to produce the evaluation along with feedback.