Artificial intelligence in hiring and workplace supervision

The fastest front to generate conflict

And the least documented. In recruitment and workplace supervision, decisions touch identified individuals, are taken daily and are almost never explained in writing. When a candidate asks why they were rejected, or an employee challenges a disciplinary measure based on a system alert, the company needs to answer with a criterion and a trail, not with the statement that this is what the tool produced.

Recruitment

Automated CV screening and test scoring are the most widespread use and the least reviewed. We work on four concrete points.

• Criterion. Which variables drive the rejection, who defined them and whether the company can explain them without referring to the vendor.
• Human review. At what moment a person actually decides, on what information, and with what record of that decision.
• Bias and non-discrimination. Whether the outcome produces effects on protected groups and what is reviewed periodically to detect it.
• Candidate data. Authorisation, stated purpose, how long the CVs of people who were not hired are kept, and what happens to that information inside the vendor's system.

The vendor is no shield either. If the tool belongs to a third party, the company still answers to the candidate and to the authority for what is done with that data, and the contract has to reflect it: what the vendor may keep, whether it can reuse candidate profiles for its other clients, and how the information is returned or deleted once the selection process ends.

Monitoring and workplace supervision

Measuring productivity, analysing internal communications or reviewing how an employee handled a customer is possible, but not unlimited. Worker data is still personal data, and the employer's power of direction has limits better fixed in advance than in the middle of a dispute.

• A concrete purpose for the measurement and proportionality between what is measured and what it is used for.
• Prior information to the worker about what is monitored, how, and with what consequences.
• Limits on personal communications, own devices and activity outside working hours.
• What evidence is kept and how it is used if it comes to support a disciplinary decision.
• Who inside the company may see monitoring results and who may not.

Rules for the employee who uses AI at work

The other side of the same problem. The employee who summarises a client contract in a free tool, uploads a database to clean it up, or generates a reply and sends it unread is taking decisions that commit the company.

• What company, client and third party information may go into an external tool, and what may not under any circumstances.
• Which deliverables require human review before leaving, and who answers for that review.
• Approved tools and the route to request approval of a new one, with a named owner.
• What gets labelled as generated or assisted by AI when the deliverable goes out to a client or to an authority.
• Consequences of breach, drafted so that they are enforceable.

The documents that have to be adjusted

None of this holds up on a memo alone. We adjust the internal work rules, the company policies, the confidentiality clauses and the data processing authorisations for candidates and employees, under Law 1581 of 2012 and Decree 1377 of 2013, so the obligation sits where it produces effects and a breach carries a consequence.

What the client receives

• A diagnosis of the AI uses currently present in human resources and workplace supervision, with the risk of each.
• The design of the human control point in recruitment and in disciplinary decisions supported by the system.
• The amendments to the internal work rules, internal policies and data processing authorisations.
• Written rules on what information may leave towards an external tool.
• Training for the human resources team and for the managers who apply these rules every day.