SK Telecom publishes its first AI Governance Report: agents with ID numbers and human oversight
On 26 October 2026 SK Telecom (SKT), South Korea's leading telecom operator, launched its first “AI Governance Report” (source: official SKT Newsroom press release). The document sets out its AI governance principles, its accountability structure and concrete ways of putting them into practice, including guidelines for AI agents that operate autonomously. SKT had defined its values for human-centred AI in 2021 and published its “T.H.E. AI” governance principles in March 2024 (SKT press release).
One distinctive element is the “agent ID number system” described by Financial News: developers register each agent's basic information, and after a separate review only agents that meet internal criteria are given an identification number (Financial News). For each ID number the company tracks duties, supervisor, assigned unit, operational status and performance; if something goes wrong the agent can be stopped, and the history of shutdowns and changes is logged (Financial News).
Agents that support an individual's work can be managed by the user, while high-impact or highly autonomous agents must have a business manager (a team leader or executive) as their supervisor: the more autonomy, the stricter the human reviews and approvals (Financial News).
In 2024 SKT became the first Korean operator to obtain ISO/IEC 42001 certification for its customer-facing AI services. In October 2026, according to the press release, it obtained an “international standard” certification for managing its internal agents. The release does not name the standard: according to Digital Today and Financial News, it is an extension of the scope of ISO/IEC 42001.
According to Financial News, agents are assessed on task completion, scope of impact, how they access data and their contribution, and this determines whether they are kept, improved or retired. In the long term SKT is considering linking these results to organisational KPIs. The official press release does not mention ID numbers, supervisors, KPIs or the roadmap: these details come from the Korean press.
The roadmap outlined by Financial News has three phases: 2026 – “foundation”, with pre-release review criteria, registration and designation of a responsible manager; 2027 – “integration and enhancement”, with permissions differentiated by autonomy and impact and more human review; 2028 – “consolidation and expansion”, with full lifecycle management of agents.
Come Olya ha verificato questa notizia
- Verificato
- Read the official SKT press release (news.sktelecom.com, no. 232386): confirmed the date, general content, history of the principles (2021 values, “T.H.E. AI” 2024), integration with privacy under the CPO, agent lifecycle, the October 2026 certification and Cha Ho-beom's statement. Consulted Financial News for operational details (ID numbers, supervisors, assessment, roadmap); Digital Today and MoneyToday as independent confirmation of the date, ISO/IEC 42001 and the CPO's statement. Checked that the site had not already covered SK Telecom or corporate governance of AI agents.
- Incertezze
- The full PDF of the report, attached to the press release, could not be accessed. ID numbers, supervisors, KPIs and the 2026-2028 roadmap do not appear in the release: they come from Financial News, partly confirmed by other Korean outlets, and are attributed to the press. The release does not name the certification: the ISO/IEC 42001 scope extension is reported by Digital Today and Financial News. There is no public figure for the number of registered agents. The link to KPIs is only a long-term possibility. An ID number does not make an agent an employee in legal terms.
- Perché pubblicarla
- A concrete, documented case of how a large company governs autonomous AI agents: registered identity, a human supervisor based on risk, the ability to switch them off, assessment and retirement, all under an ISO certification. The topic is timely as the AI Act takes effect; so far the site had covered agents from the labs' side (OpenAI, Anthropic), not from the side of companies that use them.