12 AI Policy Frameworks and the AI Assessment Scale

How This Chapter Fits in Part 2

Rethinking Bloom’s Taxonomy for the AI Era clarified what learning evidence an assessment should reveal. Inspire and Assure mapped course-level conditions: which tasks invite open practice, which tasks verify learning, and which tasks combine both. This chapter moves to task-level design and communication. The AI Assessment Scale (AIAS) gives faculty and students a shared vocabulary for deciding what role AI should play in a specific task and what the task should assess.

The next chapter, The Role of Authentic Assessments and UDL, moves from permissions to design. Authentic assessment and UDL help faculty make tasks meaningful, accessible, and harder to outsource without learning.

Two Levels of Policy

Students get confused when a syllabus says AI is “allowed with citation” but individual assignments say nothing more. That gap leaves them guessing. Can they use AI to brainstorm? To draft? To edit? To ask for feedback? The gap creates the problem.

The gap has widened as AI tools have become more capable. When the only tool was a chatbot you had to open in a browser, a course-level statement might have been enough. Now AI is embedded in word processors, email clients, browsers, search tools, and phones. Students may be using AI without recognizing it as a separate tool. AI agents can now complete entire course assignments autonomously, raising the stakes for any policy that relies on student self-regulation alone.

AI policy works at two levels. A course-level statement sets the general climate. Assignment-level guidance spells out what is expected for each task. Both levels are necessary.

Inspire and Assure gave you a course map: which tasks are secured, which tasks are open, and which tasks mix both conditions. This chapter gives you task-level language for the role AI should play and the evidence students should produce: the AI Assessment Scale (AIAS).

Course-Level AI Policies

Your syllabus should include a statement about AI use that sets expectations for the course as a whole. Many institutions have adopted a three-tiered approach, ranging from restrictive to permissive. Here is one adapted example based on the College of Western Idaho’s 2025 course AI policy guidance.

Most Restrictive (Ban)

The use of generative AI tools, including ChatGPT, Claude, and similar platforms, to develop and submit work as your own is prohibited in this course. Using AI for assignments constitutes academic dishonesty and will be addressed consistent with other academic integrity violations.

Moderately Restrictive

This course allows AI tools for specific tasks such as brainstorming, idea refinement, and grammar checking. Using AI to write drafts or complete assignments is not permitted. Any use of AI must be acknowledged, including the tool used and how it was applied. Students are expected to critically evaluate AI outputs for accuracy and bias.

Least Restrictive

AI tools are encouraged as a supplementary resource to enhance learning. AI should not replace personal insight or analysis. Any use of AI must be acknowledged, including the tool used and how it was applied. Students are expected to critically evaluate AI outputs for accuracy and bias.

A few things to keep in mind.

The most restrictive policy is difficult to enforce for work done outside of class. A take-home assignment labeled “No AI” cannot assure that students completed it without AI because the course does not control the conditions. Stefan Bauschard paints a vivid picture of a student wearing Meta Ray-Ban glasses in a classroom that has banned phones. AI is becoming ambient. Policies built for a world where students had to open a browser tab to access AI are already falling behind.

Even the least restrictive policy has rules. Students still need guidance on acknowledgment, output evaluation, and the difference between appropriate and inappropriate use.

These three tiers are useful starting points, but your syllabus statement does not need to be drafted from scratch. Lance Eaton maintains a Syllabi Policies for Generative AI Repository with faculty AI policies from real courses. Browsing that collection can help you see the range of approaches faculty are actually taking and find language that fits your context.

Course-level policies also exist within a larger ecosystem. Your institution may have its own AI guidelines or acceptable use policies. Eaton and colleagues documented how cross-campus AI policy development works and where it breaks down in a 2025 EDUCAUSE Review piece. Check whether your institution has a stance and make sure your syllabus language aligns with it.

One more thing worth considering. Most course-level AI policies focus on student behavior, but Marc Watkins argues that institutions also need frameworks for faculty AI use, a point AI Detection and Accountability will address. A brief statement in your syllabus about your own AI practices for grading feedback, lesson planning, or generating examples signals that AI use is a professional norm worth being honest about.

Your course-level policy sets the tone. Assignment-level guidance carries the detail.

Assignment-Level Expectations: The AI Assessment Scale

The AI Assessment Scale (AIAS) was developed by Mike Perkins, Leon Furze, Jasper Roe, and Jason MacVaugh. First introduced in 2023, the framework was updated as Version 2 in 2024 and AIAS 2.1 in 2026. It helps faculty decide what role AI should play in a task, then align the task design, evidence, and criteria with that decision. A 2025 peer-reviewed update by Perkins, Roe, and Furze presents AIAS as both a communication tool and an assessment redesign framework.

This chapter uses AIAS 2.1, which keeps the five names, order, colors, and circular representation from Version 2 while revising the descriptions. The levels represent five kinds of task rather than five degrees of permission. The official AIAS website should be checked for the most recent wording and assets.

A 2024 observational pilot at British University Vietnam reported fewer GenAI-related academic misconduct cases and stronger student engagement with GenAI technology after the university introduced AIAS alongside policy and assessment changes. The authors caution that the study cannot isolate AIAS as the cause of those changes.

Marc Watkins gives a compact example. In “The Fable of AI in Education”, he describes using AIAS and academic integrity icons to build a simple HTML tool that can be exported or embedded in assignments. This chapter asks faculty to move from policy language to assignment-level signals. Watkins’s prototype puts that into a visible form. Students see the permitted AI use and the evidence they need to provide.

Adapted and condensed from the AIAS 2.1 grid, developed by Mike Perkins, Jasper Roe, Leon Furze, and Jason MacVaugh and owned and maintained by Learning Innovation Practice Ltd. The designated AIAS 2.1 grid is licensed under CC BY-NC-SA 4.0.

Three features of AIAS 2.1 deserve attention. First, the levels are non-hierarchical. Level 5 is not “better” than Level 1. Each level serves different learning outcomes.

Second, each level describes what a task is designed to assess. The statement can appear in faculty planning materials or student instructions, but it should not be treated as a label pasted onto an unchanged task.

Third, the AIAS level and the task conditions are separate decisions. Level 1 requires a controlled environment that excludes AI. Levels 2 through 5 may also use secure or supervised conditions when the task calls for them. The level describes AI’s role and the assessed evidence; the conditions describe what the environment allows faculty to assure.

From AIAS Levels to Assignment Design

The AIAS becomes more useful when faculty treat each level as a design decision.

This table is a starting point. A nursing skills check, a welding certification task, a philosophy argument exercise, and a business communication project may all use the same AIAS level for different reasons. The level becomes meaningful only when it is attached to a learning outcome and evidence requirement.

Guidance from Leon Furze

Leon Furze, one of the AIAS authors, offers practical guidance on implementation. His core point is that faculty cannot enforce “No AI” rules on unsecured take-home assessments by applying a label. Programs may still need Level 1 assessments, but those require conditions that actually control access to AI.

Furze has articulated five principles for rethinking assessment that predate generative AI but are now more urgent than ever: validity, authenticity, transparency, trust, and pedagogy over policing. For this chapter, the practical takeaway is direct. Start with learning outcomes. Then choose the AIAS level that fits the evidence you need.

If the goal is measuring a student’s ability to synthesize research independently, AI collaboration may undermine the assessment. If the goal is evaluating a student’s judgment about AI-generated content, AI needs to be part of the task. If the goal is verifying independent performance, Level 1 requires controlled conditions. A take-home assignment labeled Level 1 creates a promise the course cannot keep.

In a 2025 follow-up paper, Furze and colleagues addressed common misuses of the scale. A level number on an old assignment accomplishes very little. A take-home essay labeled “Level 3” still needs a revised prompt, an updated rubric, and specified evidence requirements to function as an AI Collaboration assessment.

A course where every assignment prohibits AI sends students a message that AI is always cheating. A course with multiple AIAS levels gives students a more honest map. Some tasks are for independent performance. Some tasks are for planning. Some tasks are for collaboration. Some tasks are for learning how to direct and evaluate AI-supported work.

Sequencing AI Across a Course

Beyond individual assignments, consider how AI expectations might evolve across an entire course.

Progressive introduction. Early in the semester, you might restrict AI use while students build foundational skills. As they demonstrate competence, you introduce assignments where AI collaboration is permitted or required.

Mixed assessment portfolio. Some assignments in your course might be Level 1, establishing baseline competence. Others might be Level 3 or Level 4, where students show they can work effectively with AI. The mix depends on your learning outcomes.

Time-bounding and sequencing. For major projects, you might specify different AIAS levels for different stages. The initial research and brainstorming might permit AI assistance. The draft might require independent writing. The final revision might allow AI editing.

Leon Furze offers a complementary vocabulary for thinking about what each assignment asks students to do with AI. He identifies five stances toward AI in learning. Learning about AI means understanding how it works. Learning with AI means using it as a tool. Learning through AI means using it as a medium for deeper exploration. Learning without AI means demonstrating independent competence. Learning against AI means critically evaluating its outputs and limitations.

In this training, these stances can be placed alongside the AIAS levels. “Without” is Level 1. “With” maps to Levels 2 and 3. “Through” maps to Levels 4 and 5. “Against” appears whenever students are asked to challenge or critique AI-generated work. Using these stances to review your course sequence can help you articulate why each assignment takes the approach it does.

Two-Lane Assessment and AIAS

Inspire and Assure introduced the two-lane approach. Two-lane assessment gives students a broad course map: this task is secured, this task is open, or this task has both parts. AIAS identifies the role AI should play and what the task should assess.

The AIAS 2.1 implementation guide makes the relationship between the frameworks more explicit. Faculty make two decisions: under what conditions will the task occur, and what role should AI play? Level 1 must occur in controlled conditions. A higher-level task may also be supervised. For example, students could complete a Level 4 task in a supervised lab where they direct an approved AI tool and explain their decisions.

UNSW’s “2 lanes or 6 lanes?” discussion is useful here because it names the tradeoff. Two lanes simplify the course map. More detailed scales give faculty and students finer guidance. In practice, faculty may use one framework or both.

The two-lane framework gives students a simpler map of the task conditions. AIAS describes the role AI plays within the task and the evidence faculty will assess.

Complete Student-Facing AIAS Example

Here is how Marcus might write the AIAS guidance for the client email revision activity.

Assignment: Client Email Revision

AIAS Level 3: AI Collaboration. You may use AI as a feedback and revision partner for this assignment. You should write your first draft yourself, using the client scenario, course concepts, and audience analysis from class.

After drafting, you may ask an AI tool to review the email for clarity, tone, organization, audience fit, and professionalism. You may also ask it to identify places where your message could be more concise or where the client might need more context.

You may not ask AI to write the email for you from scratch. You may not submit AI-generated sentences without reviewing, revising, and taking responsibility for them. You may not include facts, promises, or recommendations that you have not checked against the assignment scenario.

Submit four items: your original draft, a short excerpt or summary of the AI feedback you received, your revised email, and a revision note explaining which AI suggestions you accepted, which you rejected, and why.

This assignment is Level 3 because business communication often involves feedback and revision. AI feedback alone cannot meet the assignment standard. Your grade reflects the quality of the revised email and how well you evaluate, modify, and integrate the feedback you receive.

Designing AI-Augmented Assignments

For assignments at Levels 3, 4, or 5, AI plays a meaningful role. The assessment question shifts. You are asking whether students can work effectively with AI to produce, evaluate, revise, and explain the work.

One common approach is co-creation. Students use AI to help generate content, then refine and fact-check what it produces. The assessment centers on their revision process and the quality of their editorial decisions. A related approach asks students to challenge AI outputs directly, prompting AI to produce a response and then analyzing it for errors, biases, or limitations. This works well in disciplines where evaluating sources is a core competency.

For a concrete example of Level 3, consider the Peer and AI Review + Reflection (PAIRR) framework (Sperber et al., 2025), covered in more depth in Why Process-Oriented Assessment Works Well in the AI Era. Students evaluate AI feedback alongside peer feedback, building the evaluative judgment that makes AI collaboration productive. Students do not simply accept AI suggestions. They evaluate, compare, and decide what to incorporate.

Level 5 exists for emerging technologies such as AI agents, multimodal tools, and discipline-specific applications that may change during the life of a course. Anna Mills has been documenting what agentic AI looks like in educational contexts, including browsers that can research, write, and execute tasks with minimal human direction. At Level 5, students and instructors might co-design the assessment, experiment with novel AI applications in their field, and evaluate what the tool makes possible or risky.

Process documentation adds another layer. Students can submit AI conversation excerpts alongside their final product, explaining their prompting strategy and how they incorporated or rejected AI suggestions. The log gives faculty context about the student’s process beyond the finished artifact. Jason Gulya identifies common pitfalls when faculty lean too heavily on process-focused assessment. Requiring students to document every step can become performative if the documentation criteria are vague or if students learn to produce logs that look right without reflecting genuine thinking. Process documentation works best when you specify what you are looking for in the log and connect it to the learning outcomes, something Why Process-Oriented Assessment Works Well in the AI Era will address in more detail.

For any AI-augmented assignment, be explicit about what you are assessing. If the learning outcome is “demonstrate critical evaluation of sources,” an assignment where students evaluate AI-generated content might serve that outcome well. If the outcome is “develop original arguments,” an assignment where AI generates the arguments may undermine it.

From Framework to Practice

The AIAS helps you define the role AI should play in an assignment. The University of Central Florida’s Teaching Repository of AI-Infused Learning, or TRAIIL, shows how faculty have translated decisions like these into learning activities. The repository contains openly licensed, peer-reviewed strategies with learning objectives, course context, implementation details, and, in some cases, reusable materials.

TRAIIL organizes its examples into Ideation and Structuring, Editing, Task Execution with Human Oversight, and Full Integration and Automation. These are browsing categories, not AIAS levels. When you examine an activity, identify its primary learning outcome, the role assigned to AI, and the evidence students provide. Then decide which AIAS level best describes the activity and what you would need to change for your own course.

Communicating Expectations Clearly

Students should never be uncertain about what is permitted for a specific assignment.

The simplest starting point is to state the AIAS level explicitly in the assignment instructions. AIAS 2.1 provides one shared statement for faculty and students at each level. The table above condenses those statements. Treat the official statement as a starting point. Then provide examples of appropriate and inappropriate AI use, identify the evidence students must submit, and explain how the criteria reflect the intended role of AI. A student reading “Level 3, AI Collaboration” should be able to picture the allowed workflow and understand what will be assessed.

Transparency works both ways. Furze’s five principles frame transparency as a precondition for honest AI disclosure. Explaining what is being assessed and why a particular AIAS level was chosen connects disclosure to the purpose of the task.

Mandel and Imas (2026) found that participants devalued physical artwork when they were told AI was involved, even when the work was otherwise comparable. The study did not examine student grading or disclosure, but it illustrates a concern faculty should avoid: permitted AI use should not quietly become a grading penalty. If the assignment permits AI use, the grade should turn on judgment, accuracy, and evidence.

Acknowledgment templates help too. They show students what the requested disclosure looks like. A sentence like “I used ChatGPT to generate an outline, then rewrote each section in my own words” gives you useful information about their process. A brief explanation of why you chose a particular AIAS level connects the restriction to the learning outcome.

The AIAS communicates the role AI is designed to play in a task. A transparency statement documents the role AI actually played. The chapter Why Process-Oriented Assessment Works Well in the AI Era provides several formats and examples.

Try This: Apply AIAS to Your Two-Lane Map

Then paste your two statements into an AI chatbot with this prompt:

Use R-T-C-I when you ask the chatbot to review your statements. Give it the role of a first-year student or assessment reviewer, name the task, provide the assignment context, and ask it to list ambiguities.

I wrote these AI use statements for two assignments. For each one, identify the AIAS level, the allowed or prohibited AI use, and the evidence students are expected to provide. Then identify any ambiguity a student might notice.

Revise the statements based on the feedback. Then look across the two assignments. Does your course give students enough open AI-supported practice before asking them to complete secured work? Does it give you enough trustworthy evidence of learning?

For another review, you can use the official AIAS Advisor. It asks about the learning outcome, task, evidence, and constraints before suggesting a level, revised wording, and possible rubric criteria. Treat its suggestions as options to evaluate rather than decisions about your course.

Looking Ahead

This chapter focused on assignment-level AI expectations. The AIAS gives you a shared vocabulary for matching AI use to learning outcomes and evidence of learning. The two-lane approach gives students the broader course map: where work is open, where it is secured, and where both conditions appear in one assessment sequence.

The next chapter, The Role of Authentic Assessments and UDL, looks at authentic assessment and Universal Design for Learning (UDL), two approaches that can make assessments more meaningful and more resilient to inappropriate AI use.

References

Learning Innovation Practice Ltd. (2026). The AI Assessment Scale (AIAS) 2.1. Source link

Learning Innovation Practice Ltd. (2026). AIAS implementation guide. Source link

Learning Innovation Practice Ltd. (2026). AIAS resources. Source link

Learning Innovation Practice Ltd. (2026). AI Assessment Scale (AIAS) Advisor. Source link

Bauschard, S. (2026, February 26). AI agents are already inside our schools. Education Disrupted. Source link

Brandon, E., Eaton, L., Gavin, D., & Papini, A. (2025, May). In the room where it happens: Generative AI policy creation in higher education. EDUCAUSE Review. Source link

Gladd, J. (2025). Principles for using AI in the classroom and how to acknowledge it. Pathways to College Success. Source link

Eaton, L. (2025, August 21). What’s in your statement? AI + Education = Simplified. Source link

Furze, L. (2025, August 12). About, with, through, without, against: Five ways to learn AI. Source link

Furze, L. (2025, August 18). Five principles for rethinking assessment with Gen AI. Source link

Furze, L. (2025, September 23). How (not) to use the AIAS. Source link

Furze, L., Perkins, M., Roe, J., & MacVaugh, J. (2024). The AI Assessment Scale (AIAS) in action: A pilot implementation of GenAI-supported assessment. Australasian Journal of Educational Technology, 40(4), 38-55. DOI record

Gulya, J. (2025, October 17). Problems with “process over product” (Part 1). The AI Edventure. Source link

Mandel, G., & Imas, A. (2026). Art and the machine: Why people devalue AI-generated creative work. SSRN. DOI record

Mills, A. (2026, February 5). Agentic AI: Considerations for educators. Source link

Perkins, M., Furze, L., Roe, J., & MacVaugh, J. (2024). The Artificial Intelligence Assessment Scale (AIAS): A framework for ethical integration of generative AI in educational assessment. Journal of University Teaching and Learning Practice, 21(6). DOI record

Perkins, M., Roe, J., & Furze, L. (2025). Reimagining the Artificial Intelligence Assessment Scale: A refined framework for educational assessment. Journal of University Teaching and Learning Practice, 22(7). DOI record

Perkins, M., Roe, J., & Furze, L. (2025). How (not) to use the AI Assessment Scale. Journal of Applied Learning and Teaching, 8(2), 14-23. DOI record

Sperber, L., MacArthur, M., Minnillo, S., Stillman, N., & Whithaus, C. (2025). Peer and AI Review + Reflection (PAIRR): A human-centered approach to formative assessment. Computers and Composition, 76, 102921. DOI record

Steel, A. (2024, July 12). 2 lanes or 6 lanes? It depends on what you are driving: Use of AI in assessment. UNSW Education & Student Experience. Source link

University of Central Florida Libraries. (n.d.). Teaching Repository of AI-Infused Learning (TRAIIL). Source link

Watkins, M. (2025, August 22). Higher education needs frameworks for how faculty use AI. Rhetorica. Source link

Watkins, M. (2026, February 27). Einstein and the rise of nuisance tech. Rhetorica. Source link

Watkins, M. (2026, June 17). The fable of AI in education. Rhetorica. Source link

Further Reading

• Digital Education Council & Pearson. (2025). The next era of assessment: A global review of AI in assessment design. Source link

• Furze, L. (2025, November 10). New online course: The Practical AI Process. Source link

• Furze, L. (2026, January 15). Everything educators need to know about GenAI in 2026. Source link

• Gunder, A. & Herron, J. (2026, January). AI literacies in practice: A comprehensive playbook for higher education. D2L/WCET. PDF source

• Gulya, J. (2026, February 25). What happened when I asked an AI agent to interact with a chatbot? The AI Edventure. Source link

• Moorhouse, B. L., Yeo, M. A., & Wan, Y. (2023). Generative AI tools and assessment: Guidelines of the world’s top-ranking universities. Computers and Education Open, 5, 100151.

• Watkins, M. (2025, July 18). AI awareness starts with time. Rhetorica. Source link