AI assessment tools are now embedded across hiring. What began as selective automation has quickly become part of the operational infrastructure for many HR teams. Assessments are being used at scale to evaluate skills, behaviours, cognitive ability, and role fit across increasingly large candidate pools and many of these tools now come with a ‘now with AI’ sticker slapped on top of them.
Alongside that growth has come a parallel concern: how can these systems still be trusted when candidates also have access to AI?
The anxiety among HR leaders is understandable. Candidates can now generate polished written responses in seconds, optimise CVs against job descriptions, and use LLMs during remote assessments. For recruiters and hiring teams, the concern is no longer theoretical.
Our recent report into the state of AI in talent assessments found that 43% of HR professionals are concerned about candidates using AI to cheat, while just over a quarter (26%) said they had already seen evidence of AI-assisted manipulation.
That concern is reshaping how organisations think about assessment design, fairness, and oversight.
The challenge for 2026 is not simply preventing cheating. It is protecting assessment integrity without damaging candidate trust or introducing processes so restrictive that they undermine the hiring experience itself.
The cheating problem is real, but uneven
The conversation around AI assessment tools often assumes all assessment formats are equally vulnerable. They are not.
Generative AI performs exceptionally well with text-based tasks. That makes written applications, cover letters, and open-ended written responses particularly susceptible to assistance.
Other formats, for the time being at least, are more resistant.
Our report highlights some important distinctions:
- Cognitive ability tests can be vulnerable where answers are fixed and time limits are weak
- Diagrammatic and spatial reasoning assessments are harder for AI tools to navigate in real time
- Personality questionnaires and situational judgement tests are significantly more difficult to manipulate effectively
The organisations responding most effectively are not trying to eliminate all risk. They are redesigning their assessment processes around the types of capability AI struggles to replicate on behalf of a candidate.
That includes:
- Behaviour under pressure
- Decision-making consistency
- Real-time problem solving
- Job-relevant simulations
- Stable behavioural traits
The more an assessment measures genuine behavioural patterns rather than polished written outputs, the more resistant it becomes.
Poor assessment design creates opportunity
There is a tendency to frame cheating purely as candidate behaviour. In reality, assessment structure plays a major role.
Weak assessment design creates exploitable conditions.
Common examples include:
- Unlimited completion windows
- Easily searchable questions
- Generic written tasks
- No behavioural monitoring
- Minimal variation between questions, question types
These formats were already vulnerable before generative AI became mainstream. AI has simply amplified those existing weaknesses.
Our report identifies three practical areas organisations are focusing on to reduce manipulation risk: deterrence, countermeasures, and detection.
- Deterrence
Many candidates avoid dishonest behaviour if they believe inconsistencies will be identified later in the process.
This is why reassessment stages remain effective. If candidates know they may repeat elements live or under supervised conditions, the incentive to manipulate earlier stages decreases.
Deterrence also includes transparency. Clearly communicating that assessments include behavioural analysis, validation steps, or follow-up questioning can reduce opportunistic misuse.
- Countermeasures
These are structural barriers that make external assistance more difficult.
Examples include:
- Browser restrictions
- Time-limited tasks
- Preventing copy-and-paste actions
- Randomised question sets
- Interactive assessment formats
The goal is not creating a hostile environment. Excessive restrictions often damage candidate experience and create accessibility concerns.
Effective countermeasures reduce convenience rather than attempting total control.
- Detection
Behavioural monitoring has become increasingly important within AI assessment tools.
This includes tracking:
- Frequent tab switching
- Repeated screen capture attempts
- Unusual response timing patterns
- Inconsistent behavioural data across stages
Detection works best when combined with human review. Automated flags without interpretation create false positives and unnecessary friction. Patterns matter more than isolated events.
Fairness cannot become collateral damage
One of the biggest risks in the response to cheating is overcorrection.
While heavy surveillance, intrusive monitoring, and rigid controls may reduce some forms of manipulation, they can also damage fairness and candidate trust.
Our report found that 45% of HR professionals surveyed were concerned about inaccurate or unfair evaluations linked to AI systems themselves.
This creates a difficult balance.
An organisation can build a highly restrictive assessment process and still end up with poor hiring outcomes if the system introduces bias, confusion, or irrelevant barriers.
Fairness depends on several factors:
- Relevance to the role
- Clarity of instructions
- Accessibility and neuroinclusivity
- Consistency in scoring
- Transparency around how decisions are made
If candidates do not understand what is being assessed or why, their confidence in the process can drop quickly.
Our report also highlights the importance of candidate experience in maintaining process integrity. When people perceive assessments as fair and relevant, they are less likely to justify dishonest behaviour.
Transparency is now a core requirement
Trust in AI assessment tools depends heavily on visibility into how systems operate.
Over half (58%) of respondents in our report said clear explanations of how AI is used would increase their confidence more than any other factor.
HR leaders increasingly want answers to practical questions such as:
- How is scoring generated?
- What data is being analysed?
- Where is AI influencing outcomes?
- Can decisions be audited?
This is becoming particularly important as regulatory scrutiny increases around AI in employment settings.
Opaque systems create multiple risks:
- Legal exposure
- Inability to explain decisions
- Difficulty identifying bias
- Reduced internal trust in outputs
Transparency also improves implementation quality internally. Recruiters and hiring managers are more likely to use assessment outputs appropriately when they understand the logic behind them.
Human oversight still matters
AI assessment tools can support consistency and scale. They cannot carry accountability.
Our report consistently reinforces the importance of human involvement throughout the hiring process. AI is commonly used to score, rank, and interpret results, but final responsibility still rests with people.
This oversight needs to extend beyond final hiring decisions.
Human involvement is required in:
- Psychometric test design
- Validation processes
- Monitoring fairness outcomes
- Reviewing behavioural anomalies
- Investigating flagged activity
The organisations most exposed to risk are often those treating AI outputs as inherently objective. Psychometric test platforms are only as reliable as the assumptions, data, and governance structures behind them.
The bottom line
AI assessment tools are now central to modern hiring and are not going away any time soon. Their value is real and growing, but so are the pressures surrounding them.
Preventing cheating will depend less on aggressive surveillance and more on intelligent assessment design, transparent governance, and credible human oversight.
The organisations navigating this well are approaching integrity as part of the overall assessment experience rather than a separate control layer.
A process that candidates trust is easier to defend, harder to manipulate, and more likely to identify the people actually capable of doing the role.
To see what the latest data says check out our State of AI in Talent Assessment Report here
