SK Telecom gives AI agents employee IDs, owners and a retirement path

SK Telecom published its first AI Governance Report on October 26, 2026, and it manages in-house AI agents much like staff¹. Agents that pass an internal review get an employee ID, and the company tracks each one's assigned tasks, responsible person, affiliation, operating status and performance². Agents with greater impact or autonomy must be supervised by a team leader or executive, and more autonomy brings stricter human review and approval². An internal Agent OS records execution status, usage, change history and the models and tools each agent uses, and can stop an agent when an anomaly occurs².

Seoul Economic Daily reports that SK Telecom will monitor each agent's token usage and work records, that agents judged less necessary will be suspended or retired, and that the company is reviewing how to reflect agent performance in organizational KPIs³. Its roadmap sets 2026 for building the foundation, 2027 for integrating and strengthening controls, and 2028 for a full system covering operation, evaluation and decommissioning². The Herald Business notes that SK Telecom obtained ISO/IEC 42001 certification for customer facing AI services in 2024 and extended it to its in-house agent management framework this month¹.

The useful idea is that accountability lands on a named person, scaled to how far the agent can reach. A personnel style model answers a basic question first: who owns this agent, and what may it touch. The gap is evidence: the coverage describes a framework and a roadmap, with no agent count and no measured results.

For data teams, an agent employee ID only matters if the platform can see it. That means a dedicated database role or service identity per agent, query tags that carry the agent ID, and lineage that links the tables an agent reads to the actions it takes. Token and warehouse spend per agent gives the performance review a number. Retiring an agent should look like offboarding a person: revoke grants, rotate credentials, and check that nothing downstream still depends on its output.

One step this week: start an agent register with ID, owner, purpose, autonomy level, data it can read and write, and next review date. Begin with the agents already querying your warehouse. Any row without an owner is the first one to fix.