Fair, transparent and audited: holding hiring AI to account
What an independent audit of AI in hiring should mean, the questions to ask any vendor, and how inploi answers them for both of its agents.
Holding AI in hiring to account means three things: it must be fair, which someone independent of the vendor has audited; transparent, so your team can read why it reached each recommendation; and auditable, so you can reconstruct afterwards what it did and who acted on it. A vendor that can show you all three has earned a serious conversation, and one that offers a policy document instead has not.
Why a vendor’s word is not enough
Every supplier of AI for hiring will tell you their system is fair. Most of them believe it. But fairness in a scoring model is not something you can establish by intention or by reading the code. Models pick up patterns from the data they learn from and the instructions they are given, and some of those patterns can disadvantage people for reasons that have nothing to do with the job. The only way to know is to test, and the only test a buyer should put weight on is one the vendor did not mark themselves.
The pressure to get this right is growing. Recruitment is one of the uses the EU AI Act treats as high risk, New York City already requires bias audits for automated tools used in employment decisions, and procurement, legal and risk teams at large employers now ask detailed questions before anything that touches candidates goes live. Good intentions do not answer those questions. Evidence does.
Five questions to ask any AI hiring vendor
If you are evaluating AI anywhere in your hiring process, these are the questions we would ask, and the answers we would want to hear.
1. What does it decide?
The most important answer is the narrowest one. Software that rejects candidates carries a completely different level of risk from software that ranks and recommends to a person who then decides. Ask exactly which actions the system can take without a human, and get it in writing.
2. Can we read the reasoning?
A score on its own is an assertion. A score with reasoning against stated criteria is something a recruiter can check, challenge and learn from. If the vendor cannot show you, for a real application, why the candidate ranked where they did, your team will not be able to either.
3. Who tested it, and for what?
Ask who carried out the fairness testing, whether they are independent of the vendor and what they tested for. A vendor that marked its own homework has not answered the question.
4. What is recorded?
When a candidate or a regulator asks what happened to an application, you need to be able to say what the software did, when, and which person acted on it. That needs an audit trail, not a best recollection.
5. What happens when it is wrong?
Every model is wrong sometimes. The question is whether the design catches it. A system where a human reads the reasoning before every decision catches far more than one where nobody looks until there is a complaint.
How we answer them
We built the Hiring Agent and the Candidate Agent to be able to answer these questions plainly, because they are the questions we would ask ourselves.
Neither agent rejects anyone or takes a decision on its own. The Hiring Agent scores every application against the criteria your team sets and orders the queue, as a recommendation for a person to act on. Every application stays in the queue and can be read in full: the score changes the order, not who is seen. The Candidate Agent answers candidates’ questions and may point them to roles that suit them better, but anyone can always apply anyway. Whether anyone is rejected is a decision made by a person, in the ATS you already use.
Every score comes with reasoning a recruiter can read: which requirements the candidate met, how they showed it and why they ranked where they did. Personal details are removed before the Hiring Agent scores an application, and scoring uses only the criteria your team sets. Every candidate is asked whether they are happy for AI to be used. They can say no and carry on, and neither answer helps or harms their application.
On the audit, inploi’s use of AI, including the Hiring Agent and the Candidate Agent, is independently audited for fairness and accuracy by Warden.
We chose an outside specialist precisely because the buyer should not have to take our word for it. How our AI works answers the questions buyers ask most, one by one.
A full audit trail of every agent action is available in Studio, so your team and your compliance function can see how a ranking was reached and what happened next. On the data side, inploi is ISO 27001 certified and holds Cyber Essentials Plus, candidate data is stored in the UK and EU, and a Data Processing Agreement comes as standard. Our security page sets out the detail, and documentation on our approach to AI governance is available on request through the enterprise team.
Why this matters most in volume hiring
It would be easy to think fairness testing is a concern for a handful of senior roles. The opposite is true. A frontline employer might receive tens of thousands of applications a year for roles in kitchens, shops, holiday parks and care homes. At that volume, a small systematic skew in how applications are read affects a great many real people, and it goes unnoticed precisely because nobody can read everything by hand.
That is also why doing nothing is not the safe option it appears to be. When a team cannot get through the pile, the applications that arrive first or look most familiar get attention and the rest get none. That is not neutral either; it is just untested. A system that reads every application against the same criteria, shows its reasoning and is independently audited is a better place to start than a queue nobody reaches the end of.
What to do next
Take the five questions into your next conversation with any AI vendor, including us, and see how quickly the answers come. If you would like to see how the agents score, reason and record in practice, book a demo and bring your risk and compliance colleagues along. There is more on what trustworthy AI in hiring looks like in our trust in hiring collection.
