SK Telecom gives AI agents employee IDs and human supervisors: what its first AI governance report says

SK Telecom's AI agents get employee IDs, and the ones that act most on their own get a human boss. In its first AI Governance Report, published October 26, 2026, the South Korean carrier set out how it manages the AI agents its staff build and use at work. Agents that pass an internal review receive an employee number, and any agent with greater impact or autonomy must have a team leader or executive designated to supervise it. It's one company's answer to a question that is getting harder to dodge: who answers for an agent's work, whether that agent is logging in to a customer's bank account or doing an employee's job.

This story is based on SK Telecom's October 26 newsroom release and the report itself (in Korean), SK Telecom's March 2, 2026 English press release, and reporting by Seoul Economic Daily, Financial News, The Herald Business, SBS, Edaily and RCR Wireless, checked October 27, 2026. Quotes and terms from Korean-language sources are translations.

The short version

• Employee IDs: every internal agent gets a basic ID, and those that pass a separate review and meet company criteria get an employee number, according to the report.
• Supervisors: agents with greater impact or higher autonomy must have a designated supervisor such as a team leader or executive; an agent that assists one person can be managed by that person.
• Oversight: an internal "Agent OS" records each agent's runs, token use, calls and cost, and can stop an agent's execution when something goes wrong.
• How many: the report doesn't say. SK Telecom said in March 2026 that more than 2,000 agents were in use.
• Timeline: the company plans to finish building the system in stages, from 2026 to 2028.

What does an "employee ID" for an AI agent mean in practice?

It's a registration and accountability record, not a job contract. The report describes a two-step system. First, every agent gets a basic ID so the company knows what exists. Then agents that pass a separate review and meet in-house standards are given an employee number (사번, the same kind of number human staff have). Once an agent has one, SK Telecom monitors its role, its responsible owner, its operating status and its performance. Its record also lists when it was developed and put into service, which internal systems it connects to, and the scope of its work.

The report lists three questions SK Telecom asks when deciding an agent's number and scope: whether the task has a standard operating procedure, whether its results can be measured objectively, and whether the agent can finish the work without a person stepping in. The idea of treating agents like staff isn't new for the company. On June 16, 2026, Seoul Economic Daily and SBS reported that chief executive Jung Jai-hun had presented an "AX Innovation 2.0" plan under which agents would get employee numbers, departments, job functions and permissions, and be managed like staff from hiring to departure. The October report is where SK Telecom spells out the rules.

How many AI agents does SK Telecom have, and what do they do?

The report doesn't give a number. The most recent official figure is from SK Telecom's March 2, 2026 press release from MWC in Barcelona, which said "more than 2,000 AI agents are being actively used across areas such as marketing, legal, and PR." RCR Wireless reported on June 18 that the company uses "thousands" of agents and is aiming for one agent per employee, which is a target, not the current state.

The report's examples are an HR agent, a safety and health agent, a supply-chain agent and a marketing agent, and it names A.Biz, Polaris and Playground as the internal platforms where agents are built. It doesn't describe the example agents' tasks in detail. At a September 15 marketing conference, the head of SK Telecom's AI Marketing Team described agents named T-Marketer and T-Legal: the marketing agent pulls data, sorts high- and low-performing items and writes reports, and the legal one analyzes case law, Edaily reported.

What oversight rules apply to SK Telecom's AI agents?

The more autonomous the agent, the more human checking it gets. The report sets three layers: review and approval before an agent runs, supervision while it operates, and checks afterward. Controls such as input safeguards, limits on outputting personal or confidential data, labeling of AI-generated output and logs of what the agent did are applied according to a risk assessment.

Last verified: October 27, 2026

SK Telecom also formally recognizes agents with employee numbers as a new "workforce" when it comes to measuring results. In the short term it plans to link agents' evaluation results to executive evaluation and pay and to rewards for staff, according to the report, and in the longer term it is considering adding agent performance to organizational KPIs. Anthropic has published its own attempt to measure agent work against human work, which found agents running six times as many hours as its employees by August 2026; SK Telecom hasn't published comparable numbers.

What is SK Telecom's timeline?

The report's roadmap runs in three steps. In 2026, the company sets standards for identifying and registering agents, naming their owners and limiting their permissions and data access. In 2027, it varies permissions and controls by autonomy and impact, and strengthens human review, approval and intervention. By 2028, it aims to have rules covering each agent's operation, evaluation, maintenance, upgrades and retirement in place across the company.

In October 2026, SK Telecom extended its ISO/IEC 42001 certification, the international standard for AI management systems, to cover its internal AI agent management. It first received the certification in 2024 for customer-facing AI services, saying it was the first Korean telecom operator to do so. Chief privacy officer Cha Ho-beom, who oversees AI governance, said in the release that "the importance of AI governance is growing even greater in the age of AI agents," in The Herald Business's translation.

Why it matters

Most companies deploying agents haven't said who is accountable for them. SK Telecom's answer is a named human for every high-impact agent, plus logs and a kill switch. That addresses the risk behind the case against unattended agents: agents running for long stretches with nobody approving each step. What the report doesn't settle is what happens when a supervised agent still gets something wrong. RCR Wireless raised that question in June: whether responsibility falls on the agent's human manager, the governance team or the model vendor. The report assigns supervisors but doesn't answer it.

What's confirmed, reported and uncertain

• Confirmed (SK Telecom): the October 26, 2026 report and release; employee numbers for agents that pass review; team-leader or executive supervisors for high-impact or highly autonomous agents; the Agent OS logging and stop function; the evaluation and retirement rules; the 2026–2028 roadmap; the October 2026 ISO/IEC 42001 extension; more than 2,000 agents in use as of March 2, 2026.
• Reported: the June 16 AX Innovation 2.0 announcement (Seoul Economic Daily, SBS); "thousands" of agents and the one-agent-per-employee goal (RCR Wireless); T-Marketer and T-Legal (Edaily).
• Uncertain: how many agents have employee numbers today; whether any agent has been retired under the rules; who is liable when a supervised agent errs.