Five AI Questions Foreign Headquarters Ask

In our last article we said that HR and labor at a foreign-invested company has one more variable: headquarters.

Today we look at a new question that this variable has recently created.

Only a few years ago, headquarters did not ask the Korean entity’s advisors questions like this. Now it does. AI items appear on vendor assessment forms, and headquarters’ legal teams want to confirm how advisors use AI.

The reason is simple. Regulations in the country where headquarters is located flow down to the Korean entity.

Europe — AI used in hiring and evaluation is “high-risk”

The EU Artificial Intelligence Act (AI Act) classifies AI used for recruitment, screening of candidates, performance evaluation, task allocation, monitoring of workers, and promotion and termination as high-risk. The HR domain is included in its entirety.

What high-risk classification requires is not a ban on use but management obligations: inform workers; ensure that people can meaningfully oversee and override the system; periodically check for discriminatory outcomes and keep records; and be ready to explain automated decisions.

One point needs attention here. The date on which these high-risk obligations take effect has recently been pushed back.

It was originally 2 August 2026, but through the so-called Digital Omnibus the EU postponed the obligations for stand-alone high-risk AI falling under Annex III to 2 December 2027. The European Parliament gave final approval on 16 June 2026 and the Council on 29 June.

Not everything has been postponed, however. The transparency obligations in Article 50 still apply from 2 August 2026. And there is a fact that matters more in practice: a postponed deadline does not mean headquarters stops acting. Headquarters that have already set internal standards demand them of their suppliers and advisors first. Headquarters’ requirements arrive before the regulations do.

United States — it varies by state, but the direction is the same

In Illinois, an amendment to the Human Rights Act (HB 3773) has been in force since 1 January 2026. Employers using AI for hiring, promotion, discipline, discharge and similar decisions must notify employees and applicants, and if that AI produces discriminatory outcomes against protected classes, lack of intent is no defense. Using zip codes as a proxy for protected characteristics is also prohibited.

Since July 2023, New York City has required automated employment decision tools (AEDTs) to undergo an independent bias audit once a year, a summary of the results to be published on the employer’s website, and candidates to be notified at least 10 business days in advance.

Colorado passed the first comprehensive AI law in the United States (SB 24-205) in 2024, but after two postponements its effective date has been pushed to 1 January 2027 and its content considerably narrowed.

Each state’s circumstances differ. Yet the direction of regulation converges on one point.

Korea — already in force

Korea’s Framework Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trust is already in force (the current act took effect on 21 July 2026).

A business that provides services using generative AI must notify users in advance and indicate that outputs were generated by AI. If the system is high-impact AI—AI that may significantly affect human life, physical safety or fundamental rights—the act further requires risk-management measures, explanation measures, user-protection measures, human management and oversight, and the preparation and retention of documents.

In short, HR managers at foreign-invested companies now look at three layers of standards at the same time: the regulations of the headquarters’ home country, headquarters’ own internal standards, and Korean law.

Five questions to ask your advisor

The questions that actually come up in headquarters audits and vendor assessments narrow down to five. You may use them as they are when selecting a new advisor or renewing a contract.

1. Was the conclusion reached by AI or by a person?

If AI produces the grades and determinations, the same matter may receive different answers, and the reasoning cannot be traced back.

2. Is our company’s data used to train AI models?

If materials handed over during the advisory process are used for model training, they cannot be retrieved.

3. Are sentences written by AI labeled?

Without a label, you cannot tell where the expert’s judgment ends.

4. Who is the responsible person?

“The company manages it” is not an answer. There must be a name and contact details.

5. Is there a record?

What a headquarters audit asks for is not the result but on what basis, when, and what was decided.

FAIR HR & Labor Consulting’s answer

We have published our answers to these five questions on our website as guidelines. Rather than explaining them verbally, we decided it was right to put them in a document you can check at any time.

• FAIR AI Usage Guidelines (in Korean)

The gist is fivefold.

First, rules make the determination and AI provides the explanation.

Diagnostic grades and risk determinations come from rules designed by certified labor attorneys. AI turns those conclusions into readable text and finds and attaches supporting materials. That is why entering the same answers twice produces the same conclusion.

Second, we do not train models on client data.

Instead of training models on our data, we look up and read materials when needed and attach the sources. Clients’ materials do not remain in the model.

Third, sentences generated by AI carry an AI label.

On diagnosis and consultation screens, users are told before they begin that AI is used in an assisting role.

Fourth, we have disclosed the person responsible for management and oversight.

Their name and contact details are written in the guidelines.

Fifth, we have also disclosed where we do not use AI.

For example, we do not use generative AI in the performance management system dedicated to a client company. Having said so publicly, we judged that hiding the button on screen was not enough, so we blocked the function itself on the server.

Finally

The era in which not using AI was the safe choice is over. Conversely, using it without disclosure has become the risk.

If you work at the Korean entity of a foreign-invested company, this issue will come down from headquarters sooner or later. If the advisor then answers, “We do not use AI,” headquarters will ask again: whether that is really true, and if AI is used, how it is managed.

FAIR HR & Labor Consulting has advised foreign-invested companies, including Microsoft, GE and CITIBANK, on HR and labor matters. If you have a matter in progress, we will start by telling you what to check first.

• FAIR AI Usage Guidelines (in Korean)
• Foreign Company Support Center
• Request a consultation (in Korean)