JIPYONG LLC

Jipyong News|KOREA LEGAL INSIGHT

AI-Based Safety Management Systems and the Duty to Ensure Safety and Health under Korea’s Serious Accidents Punishment Act

2026.08.18

Introduction
<div> 
The increasing adoption of AI in workplace safety management has raised an important question under Korea’s Serious Accidents Punishment Act (the “SAPA”): to what extent can the introduction and operation of AI-based safety management systems demonstrate that a company has fulfilled its statutory duty to establish and implement an effective safety and health management system? 
 
The issue has gained significance following a recent decision of the Suwon High Court (Suwon High Court Decision No. 2025No1502, decided April 22, 2026).  The court held that liability under the SAPA may arise where a serious accident results from a breach of duties relating to the establishment and implementation of a safety and health management system, even without establishing a breach of a specific duty under the Occupational Safety and Health Act.  This article examines the legal framework and practical considerations relevant to the implementation of AI-based safety management systems.</div>
 
 
1. Relevant Legal Frameworks 
 
A) The SAPA (Duty to Establish and Implement a Safety and Health Management System)

<div> 
Article 4(1) of the SAPA and its Enforcement Decree require employers and responsible managing officers to establish a safety and health management system.  Although AI-based risk assessment systems may assist in identifying hazards and reviewing improvement measures, they do not relieve employers of their obligation to establish the underlying work procedures required by the Enforcement Decree.</div>
 
B) The Framework Act on Artificial Intelligence (Obligations of AI-Using Business Operators)

<div> 
The Framework Act on Artificial Intelligence entered into force on January 22, 2026.1)  Although AI-based safety management is not expressly designated as high-impact AI, certain systems used for workplace safety may constitute high-impact AI subject to the Act’s obligations, including risk management and human oversight requirements.  An employer that merely uses an AI safety management system supplied by an external vendor solely for its own workplace is likely to be treated as an artificial intelligence user rather than an artificial intelligence-using business operator and therefore are not subject to these provisions.  However, employers that develop, commission, or provide AI systems to third parties may themselves be regarded as artificial intelligence-using business operators and be subject to the Act.  The Ministry of Science and ICT’s Guidelines on the Obligations of High-Impact AI Business Operators further emphasize meaningful human oversight, including the ability to understand AI outputs, intervene where necessary, and take corrective action.</div>
 
C) EU AI Act (Human Oversight Obligations)

<div> 
The EU AI Act entered into force in August 2024, although provisions imposing obligations on high-risk AI systems will apply from December 2027.2)  While AI-based safety management systems used in ordinary manufacturing or construction workplaces would not ordinarily fall within the Act’s high-risk classification, certain applications—such as those used for safety management in critical infrastructure or for monitoring and evaluating worker behavior—may be subject to its requirements.  In any event, the Act’s human oversight requirements provide a useful reference point by emphasizing that AI systems should remain subject to meaningful human supervision and intervention.</div>
 
 
2. Practical Considerations 
 
A) Introduction of an AI System May Not, by Itself, Satisfy Statutory Duty

<div> 
The mere introduction of an AI-based safety management system is unlikely, in itself, to be regarded as constituting performance of the relevant duty to establish and implement a safety and health management system.  It is important to recognize that an AI system is merely an auxiliary tool that may assist in the performance of statutory duties, rather than a substitute for those duties themselves. 
 
Indeed, in some circumstances, introducing an AI system may operate as a factor aggravating liability rather than mitigating it.  Particular caution should be exercised where: 
 
<table border="0" cellpadding="0" cellspacing="0">
	<tbody>
		<tr>
			<td valign="top" width="3%">  </td>
			<td width="97%">(i) the introduction of an AI system is accompanied by a reduction in, or reassignment of, personnel responsible for workplace safety management;</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(ii) budget is allocated for the introduction of the AI system, while the budget for training safety managers and supervisory personnel is reduced;</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(iii) no manual or protocol is established to govern the human response to AI-generated warnings or alerts;</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(iv) following the introduction of the AI system, the workplace’s safety and health objectives and management policies are not revised to reflect the operation of the system.</td>
		</tr>
	</tbody>
</table>
</div>
 
B) AI System Malfunctions May Not Necessarily Shift Responsibility

<div> 
Because AI-based safety management systems are inherently based on statistical models, it is impossible to eliminate malfunctions completely, including false positives and false negatives.  Responsibility for a serious accident resulting from an AI system malfunction will depend largely on whether the employer appropriately recognized the system’s limitations and maintained effective human oversight.  Where those limitations have been adequately assessed at the implementation stage, expressly reflected in workplace safety and health management rules, and accompanied by appropriate human oversight and inspection systems, a false negative alone is unlikely to be attributed to the responsible managing officer.  Conversely, where human oversight has been reduced without first assessing the system’s limitations, or where no procedures exist for responding to identified limitations, an accident resulting from a false negative is more likely to be regarded as the consequence of a systemic failure.</div>
 
C) Outsourcing AI Systems May Not Relieve Employers of Their Statutory Duty

<div> 
In the Suwon High Court Decision 2025No1502, the court held that the mere fact that the risk assessment under the Occupational Safety and Health Act had been outsourced to an external consulting firm did not relieve the employer of its duty to establish the relevant work procedures.  As most AI-based safety management systems are introduced through external solution providers, and their operation and maintenance are often entrusted to those providers, the same principle may apply where an AI-based safety management system is outsourced. 
 
In particular, the following matters are likely to remain the responsibility of the responsible managing officer: 
 
<table border="0" cellpadding="0" cellspacing="0">
	<tbody>
		<tr>
			<td valign="top" width="3%">  </td>
			<td width="97%">(i) determining which hazards and risk factors within the workplace the AI system should be configured to detect;</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(ii) establishing procedures for human responses to hazards or risks detected by the AI system; and</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(iii) supervising and evaluating the limitations of the AI system and any blind spots in its operation.</td>
		</tr>
	</tbody>
</table>
</div>
 
D) Ensure Meaningful Human Oversight of AI Operations

<div> 
The principal value of an AI-based safety management system lies in its ability to identify warning signs before an accident occurs and communicate them through the workplace safety and health management system.  However, even where an AI system identifies hazards or risk factors, the benefits of introducing the system will be undermined if there is no meaningful human response or if such response exists only in form.  Accordingly, when implementing an AI-based safety management system, employers should establish in advance and document: 
 
<table border="0" cellpadding="0" cellspacing="0">
	<tbody>
		<tr>
			<td valign="top" width="3%">  </td>
			<td width="97%">(i) the classification of AI-generated alerts according to their level of risk and the corresponding response procedures;</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(ii) the personnel responsible for responding at each alert level (e.g., frontline workers, supervisors, safety managers, and responsible managing officers);</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(iii) the response criteria applicable to each alert level; and</td>
		</tr>
		<tr>
			<td valign="top">  </td>
			<td>(iv) procedures for recording and reporting responses.</td>
		</tr>
	</tbody>
</table>
 
Employers should also strengthen the ability of key personnel within the workplace safety and health management system—such as safety managers and supervisors—to critically evaluate AI systems through regular training.  Such training should cover, among other things, the operating principles of the AI system, its limitations, the possibility of false positives and false negatives, and variations in system performance under different environmental conditions.  Consideration may also be given to incorporating into workplace policies the principles of human oversight set out in Article 14(4) of the EU AI Act, including the authority to refrain from using the AI system or to disregard or override its outputs, and the authority to safely stop the operation of the AI system.</div>
 
 
Conclusion

<div> 
AI-based safety management systems can serve as an effective means of supporting the establishment and implementation of a workplace safety and health management system.  At the same time, however, the introduction of AI systems gives rise to a new dimension of compliance obligations.  This is because the obligations of AI business operators under the Framework Act on Artificial Intelligence, the obligation of human management and oversight set out in the Ministry of Science and ICT’s Guidelines, and the human oversight obligation under the EU AI Act all require human intervention in the operation of AI systems and the assumption of responsibility by humans.  As regulatory guidance and practical experience continue to develop, the legal and practical issues surrounding AI-based safety management systems are expected to become more refined. 
 </div>
<a href="https://www.jipyong.com/kr/board/news_view.php?seq=15374" target="_blank">[Link to Korean Newsletter]</a> 
 
<hr align="left" background-color="#cc0000" size="1" width="33%" />
<div style="margin-bottom: 8px; font-size: 16px; line-height: normal;">1) The Framework Act on the Development of Artificial Intelligence and Establishment of a Foundation for Trust (Act No. 20676) (the “Framework Act on Artificial Intelligence”).</div>

<div style="margin-bottom: 8px; font-size: 16px; line-height: normal;">2) The Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (the “EU AI Act”).</div>

</div>

KOREA LEGAL INSIGHT Preparation for the upcoming amendment to Korea’s Personal Information Protection Act 2026.09.07

GLOBAL LEGAL INSIGHT Vietnam’s Resolution 66.18: Temporary Streamlining of Administrative Procedures and Business Conditions 2026.08.13

Back to List

JIPYONG LLC

Related Practices

Professionals

Related Practices

Professionals

SHARE

Disclaimer

Limited Liabilities