AI Changed How People Work. Should It Change How We Assess Them?

Picture two assessments on the same afternoon.

In the first, a cloud engineer sits a professional certification. The environment is locked down: no browser, no documentation, no AI. She has 90 minutes to configure a network and secure a set of hosts using only what she knows.

In the second, a developer completes a hiring assessment. He has search, documentation, and an AI coding assistant open, exactly as he would on his first day at work. The task is to find and fix a production bug.

Both assessments are well designed. They just answer different questions.

There isn’t one right AI policy. There’s a right policy for what your assessment needs to prove. If a credential certifies that someone has mastered a body of knowledge or can work unaided, restricting outside tools makes sense. If a hiring assessment needs to show how someone actually works, giving them the same resources they’d have on the job may tell you more. The mistake is choosing a policy before deciding what you’re measuring, or choosing a format that leaves you no choice at all.

What Is AI-Assisted Assessment?

An AI-assisted assessment lets candidates use AI tools, such as ChatGPT, Claude, or GitHub Copilot, while they complete the tasks being scored. Two terms often get mixed up with it:

• AI-scored assessment: AI grades the candidate’s work.
• AI-proctored assessment: AI monitors the session for integrity.

AI-assisted assessment is about the candidate’s toolkit, not the grader or the proctor. And it’s rarely all or nothing. Between “no outside resources” and “everything you’d have on the job” sit options like documentation only, or search without AI. Each setting measures something a little different.

The Pressure Behind the Question

Three numbers explain why AI policy has landed on so many agendas this year.

75%. The share of knowledge workers already using AI at work, according to the 2024 Work Trend Index from Microsoft and LinkedIn. The same research found two-thirds of leaders wouldn’t hire someone without AI skills.

39%. The share of workers’ core skills employers expect to change by 2030, per the World Economic Forum’s Future of Jobs Report 2025. AI and big data top its list of fastest-growing skills.

60%. The share of U.S. hiring professionals who say AI is making it harder to evaluate candidates’ real skills, from Western Governors University’s 2026 Workforce Decoded survey. The share still working out how to evaluate AI skills rose from 16% to 32%. The same survey found employers leaning harder on credentials: 91% called certificates valuable skill indicators, up from 86% the year before. That raises the stakes for what a credential actually proves.

Education researchers are asking the same question. Writing for AQA, UCL professor Rose Luckin argues that AI should change what we test, not just how we police it, and points out the paradox of preparing people for an AI-enabled world while assessing them in conditions where AI is forbidden. The way through starts with a question you already know how to answer: what should a passing result prove?

Not sure what your assessment policies should look like in the age of AI? We’d love to help you find the right approach. Book an introductory call to talk through your program’s needs.

Matching the Policy to the Purpose

Policy	What a pass proves	Best fit	What it takes to deliver well
AI restricted	The candidate can perform unaided and has mastered a defined body of knowledge	Professional and foundational certifications, regulated credentials	Isolated environments, ID verification, proctoring, session recording
AI allowed	How the candidate works with the resources they’d have on the job	Hiring assessments, role simulations, practitioner evaluations	One consistent set of permitted tools for everyone, tasks that reveal judgment, outcome-based scoring
AI as the skill	The candidate can build, deploy, or troubleshoot AI-powered systems	AI certifications, AI engineering roles, enablement for AI products	Environments provisioned with the AI services the task needs, scored on the working result

Notice what each row depends on: control over what’s inside the assessment. Knowledge questions are an efficient way to confirm what someone knows, so they often play a role in the first row. But proving someone can perform unaided, work with real tools, or build with AI takes hands-on tasks in an environment where you decide what’s available. That’s where performance-based assessment comes in. It can support all three policies, because you design the environment around what you want to measure.

Seven Steps to an AI Policy You Can Defend

1. Start with the job, not the tool. Write down what the role really involves and the conditions it happens in. A network engineer restoring service during an outage may have no internet. A developer shipping features has every resource available.
2. Decide which tasks must be done unaided and which reflect real-world workflow. Most roles include both. Sort each task you plan to assess into one of the two groups before you think about tools.
3. Tie the policy to the decision the result supports. A credential validates defined competencies, and some require candidates to demonstrate mastery without outside assistance. That’s the line the Institute and Faculty of Actuaries draws: its September 2026 assessment regulations treat any use of generative AI tools as inappropriate conduct. A hiring score informs “will this person do well on our team,” which often points to access. Canva took that approach in 2025, expecting engineering candidates to use AI tools in technical interviews while still evaluating their engineering fundamentals.
4. Choose a format flexible enough to carry it. If your current exam is mostly multiple-choice, our guide to converting an MCQ exam to a performance-based assessment shows how to add hands-on tasks. A blended assessment keeps knowledge questions for breadth and adds performance tasks for depth.
5. Make whatever policy you choose enforceable. For restricted exams, that means isolation, ID checks, proctoring, and recording, all supported on TrueAbility’s certification platform. For AI-allowed assessments, it means a defined toolset and the ability to review how candidates worked, not only what they submitted.
6. When AI is the competency, assess it hands-on. Using AI well and building with AI are increasingly distinct from the underlying technical skill, so decide which you’re measuring, or both. Assess AI fluency by observing how candidates work with permitted tools on realistic tasks. Assess AI engineering by having candidates deploy, integrate, and troubleshoot AI workflows in a working environment, the approach behind performance-based AI certification.
7. Publish the policy and review it every year. State what’s permitted, why, and how it’s enforced, then revisit it as the role evolves.

Common Missteps

• Debating “AI or no AI” in the abstract. Without agreement on what a pass should prove, the discussion goes in circles.
• Letting the format make the decision. If your exam can’t change what candidates have access to, you haven’t chosen a policy; you’ve inherited one.
• Allowing AI with no defined toolset. Different tools and versions make scores harder to compare.
• Restricting what the job never restricts. In hiring, an unrealistic setup can filter out people who’d thrive in the role.
• Keeping the rules unwritten. Candidates trust a process they understand. Our post on the ethics of AI-assisted hiring and skill assessment explores why transparency matters.

The Bottom Line

AI should change how you assess people, but not in the same way for every assessment. Decide what the result needs to prove, then design the environment, the permitted tools, and the controls to match. A format that gives you that flexibility is worth more than any single policy.

Organizations including Google, Elastic, Canonical, and the Casualty Actuarial Society use TrueAbility to deliver hands-on assessments. See how our configurable assessment environments adapt to whatever your assessment needs to prove, across certification, training, and hiring. Book a demo.

FAQ

What does AI-assisted assessment mean? An assessment where candidates may use AI tools such as ChatGPT, Claude, or Copilot while completing scored tasks. It’s separate from AI-scored assessment (AI grades the work) and AI-proctored assessment (AI monitors the session).

Should candidates be allowed to use AI during an assessment? It depends on what the result needs to prove. If a credential certifies mastery of a body of knowledge or the ability to work unaided, restricting outside tools makes sense. If a hiring assessment needs to show how someone actually works, giving them the resources they’d have on the job may tell you more.

Is AI-assisted assessment only for hiring? No. Practitioner-level certifications and training programs can choose to use it too when the role they’re assessing involves working with AI tools on the job.

What’s the difference between allowing AI and testing AI skills? Allowing AI means candidates can use tools like ChatGPT, Claude, or GitHub Copilot to help complete a task, such as debugging code. The assessment still measures their underlying technical skills. Testing AI skills means evaluating AI proficiency directly, whether that’s writing effective prompts, identifying errors in AI-generated output, or building, deploying, and troubleshooting AI-powered systems.

How do you keep scores fair when AI is allowed? Give every candidate the same environment and the same permitted tools, and score the outcome of the task. Session review can then show how candidates got there, which is useful context when you’re evaluating judgment.

Can a single platform support restricted and AI-allowed assessments? A performance-based platform can, because the environment is configurable. With TrueAbility, internet access can be enabled, specific applications and tools can be included, or candidates can be isolated from outside resources, depending on what you want to measure. Elastic’s performance-based certification exams, for example, give candidates access to Elastic’s documentation inside the exam environment while blocking all other websites, including Google and Stack Overflow.

Do multiple-choice questions still have a place? Yes. They’re an efficient way to confirm breadth of knowledge, and many programs pair them with hands-on tasks in a blended format.

Which employers already allow AI in interviews? Canva began expecting engineering candidates to use AI tools in technical interviews in 2025, and says it’s looking for strong engineering fundamentals amplified by AI. Meta confirmed the same year that it was testing AI-assisted coding interviews.

What’s the first step toward an AI policy? Pick one assessment and write down what a passing result should prove. Then talk with the TrueAbility team about designing the environment to match.