AI News
SK Telecom Gives Its AI Agents Employee IDs and Performance Reviews
South Korea's SK Telecom published its first AI Governance Report on October 26, and its most striking idea is also its simplest: treat AI agents like employees. The carrier assigns employee ID numbers to AI agents just as it does to human staff, and continuously monitors each agent's role, designated supervisor, operating status and performance. Agents with greater impact or higher autonomy must be supervised by a team leader or executive, not a rank-and-file employee. The company says it is formally recognizing AI agents as a new form of labour, and reviewing plans to reflect their performance in the key performance indicators of its organizations. Agents judged less necessary on evaluation are suspended or retired. A telecom company doing HR for software sounds like a gimmick. It is, in fact, the most practical AI governance story of the month — because the alternative, ungoverned agents, is already costing companies real money.
What the system actually tracks
The report covers SK Telecom's AI governance principles, its accountability framework, specific action plans and results, produced to strengthen what the company calls "AI trust management" — an approach aimed at improving the safety and reliability of AI technology. For in-house agents, the company operates an agent operating system, called AXMS, that manages agents developed and run in various environments under consistent standards. Each agent's token usage and work records are examined, and human review and approval procedures are tightened for agents with greater autonomy. The company also plans to recognize agents issued employee ID numbers as an officially new workforce and to run a system linking their operating performance to organizational performance. "We are reviewing ways to share agents' performance with our employees," the company said. "Over the longer term, we will consider reflecting it in organizational KPIs." The framework did not appear overnight: SK Telecom established human-centred "AI pursuit values" in 2021, unveiled AI governance principles known as "T.H.E. AI" in March 2024, and announced plans in June 2026 for a governance framework assigning agents an employee number, organizational affiliation, job function and authority level, with managed access to data and security systems.
Why companies are suddenly doing this
The scale of the problem arrived quietly. Microsoft said on October 21 that Volkswagen Group now has more than 330,000 employees on Copilot, with over 26,000 agents active across the group in the past month. Twenty-six thousand agents is a workforce, and workforces need management: who is allowed to do what, who checks the work, who answers when something breaks. A Redwood Software survey of 300 senior technology, risk and AI leaders, published October 22, found that 77% said their company had already suffered a negative consequence from an ungoverned AI agent. Only 42% keep a central registry of the agents they have deployed, and only 51% can monitor agent activity in real time. The most common consequence was unexpected cloud costs, followed by operational disruption and compliance issues. Most companies, in other words, bought agents first and discovered the management gap later. SK Telecom's approach reverses that order: identity, supervision and retirement rules before scale, not after.
What to copy, and what to be sceptical about
The parts worth copying are the boring ones. Give every agent an identity, a named human supervisor whose seniority matches the agent's autonomy, a record of what it did and what it cost, and a retirement path for agents that stop earning their keep. Tighten human approval as autonomy rises, rather than treating all agents the same. Those are ordinary management practices applied to a new kind of worker, and they work precisely because they are ordinary. The parts to be sceptical about are the numbers that sound like results but are not. SK Telecom's report also leans on external validation: it became the first domestic telecom operator to obtain ISO/IEC 42001 certification — the international standard for AI management systems — for customer-facing AI services in 2024, and secured the same certification for its in-house agent management framework this October. Certification is a real signal, but it certifies that a management system exists, not that every agent behaves. The company also disclosed that its information protection investment last year was 111.1 billion won, up 70.3% on the previous year, with a long-term roadmap to invest 700 billion won in information protection and AI safety over five years. Spending is not a safety result either. The honest reading of SK Telecom's report is that it is a template, not a trophy: a large company showing, in public, how it intends to keep track of software that acts on its behalf. As chief privacy officer Cha Ho-beom put it, "as the era of AI agents fully unfolds, the importance of AI governance is greater than ever." The companies that will learn that lesson the hard way are the ones with thousands of agents and no list of what they are.