Responsible AI3 min read
“Human in the loop” is not a slogan. The rules we built into Talenda’s AI
Every AI hiring vendor promises a human in the loop. Here are the concrete rules Talenda uses so AI suggests and people decide, with confirmations and logs.
The pain point
“Our managers like the idea of AI in hiring, but nobody can tell me what it is actually allowed to do on its own.”
Key takeaways
- “Human in the loop” only means something when you can point to the exact steps where a person must act.
- AI should never make hiring, rejection or offer decisions, and its suggestions should carry reasons people can check.
- Actions that change or delete records should need an explicit human confirmation of a server-prepared draft. A typed “yes” to a chatbot is not enough.
- Every AI call should be logged so that actions can be traced and audited.
Read ten AI recruitment websites and nine will say “a human stays in the loop”. I have written that sentence myself. But when HR leaders ask me what it means in practice, the honest answer has to be specific: which steps, which buttons, which logs.
Trust is the real constraint on AI in hiring. Gartner’s research, reported by HR Dive, found that only about a quarter of candidates trust AI to evaluate them fairly. Hiring teams feel a version of the same doubt. So here are the rules we actually built into Talenda, written down so you can hold us to them.
Rule 1: AI suggests. People decide.
Talenda does not hire, reject, advance or make offers to candidates on its own. AI structures CVs, suggests matches, runs screening conversations and summarises them. Decisions with a significant effect on a person, including hiring, rejection, offers and HR changes, are made by people.
Rule 2: Every suggestion shows its reasons
A match without a reason is a black box. In Talenda, suggested candidates come with match reasons that a recruiter can read and check against the full profile. When the reasoning doesn’t hold up, the recruiter can see it immediately and move on. (I explain why this matters in why keyword filters miss your best candidates.)
Rule 3: Changes need a draft and a Confirm button
Talenda’s assistant, Ask Anything, can do real work: create a job, update a candidate profile, arrange an interview, delete a duplicate application. That is useful, and it is exactly where AI needs a guardrail.
So the assistant never applies those changes directly. It prepares a draft on the server and shows you exactly what will change, including the records a deletion would affect. Nothing happens until a person clicks Confirm. Drafts expire, and are rejected if the record has changed in the meantime. Typing “yes” in the chat does not count, and neither does anything the model says. Only the human confirmation does.
Rule 4: AI works with your permissions, never more
Every AI action runs with the permissions of the signed-in user and stays inside that organisation’s workspace. The AI never gets administrator or system rights. If you can’t see a record, neither can the assistant working for you.
Rule 5: Answers come from your records
When the assistant answers a question about a candidate, it reads the saved profile. If the profile doesn’t contain the answer, the answer is “unknown”. Staff can review and correct AI-generated profiles and summaries.
Rule 6: Everything is traceable
Talenda logs each AI call, including the feature, model, time, organisation and user, along with agent tasks, messages and changes to HR records. When someone asks “why did this happen?”, there is a record to answer from.
Questions to ask any AI hiring vendor
- Which actions can your AI take without a person confirming them?
- Can I see the reason behind every candidate suggestion or score?
- Does the AI ever have more access than the user it is working for?
- Is our data used to train models, or shared with other customers?
- What is logged, and can we audit it?
If a vendor can’t answer those clearly, “human in the loop” is a slogan. You can read how we answer them, in full, in our privacy policy. For the data-protection side, see my data privacy checklist for AI in recruitment.