SK Telecom Opens Up AI Safety Testing, Calls for Fewer Duplicated Country-by-Country Checks
The carrier has published an AI governance report and is building a framework in which outside companies and universities help probe its systems for risk. Its argument: safety tests repeated market by market hold back both corporate expansion abroad and user choice.
SK Telecom says it has published an “AI Governance Report” laying out the principles, operating practices and results behind how it manages AI safety and accountability. Alongside the risk reviews it already runs internally, the company is also moving toward joint verification involving outside firms and universities. Easing the burden created by differing national AI rules, the company argued in a newsroom post, requires that testing methods and results be mutually intelligible and comparable.
Why do differing national AI rules matter?
SK Telecom notes that as countries race to secure AI capabilities, divergence and overlap in regulation are both growing. Even for the same AI customer-service product, if each market demands different tests or report formats, a company ends up repeating much the same work. The cost of demonstrating safety, the company says, becomes a drag on overseas expansion.
That burden also affects what users can choose from, the company argued. Large firms with ample staff and capital can absorb repeated testing, but for startups and mid-sized companies it becomes a barrier to entry. When it is harder for new players to enter a market, the range of services available to users shrinks as well. Rules intended to protect AI sovereignty, the company suggests, may end up deepening dependence on a particular system.
Can test results from another country simply be reused?
The starting point SK Telecom proposes is making testing records comparable across borders. That means documenting, together, which AI model was tested, which risks were examined, what tools and conditions were used, and what limitations apply to the results. The idea is to let each country keep its own criteria for judgment while still being able to understand what another body actually verified.
The company also pointed to examples of international joint testing. SK Telecom said the International Network of AI Safety Institutes, launched in 2024, ran joint tests in 2025 on sensitive-information leakage, fraud and cybersecurity risks, comparing evaluation methods in the process. Still, it stressed that moving from merely comparing results to actually reducing duplicated testing will require intergovernmental agreement on how far results from other countries can be recognized.
How are the company’s domestic services checked?
SK Telecom says it screens for risk by combining self-assessments by business teams with reviews by a dedicated specialist organization. Oversight runs through every stage from service planning to operations, and the company uses SPeCTRA, a quality-evaluation platform in which AI grades the answers produced by other AI, to examine safety and accuracy. The report covers these management principles, accountability structures and implementation plans.
A cybersecurity council for homegrown AI, launched in October, will form one part of the company’s external verification efforts. According to the company, participants include six information-security firms and three universities. The plan is to run tests in which SK Telecom’s A.X K and Upstage’s Solar attack each other to expose weaknesses, with universities designing and evaluating the tests from an arm’s-length position.
AI agents, which carry out tasks on a user’s behalf, are also in scope. SK Telecom said that within the “AI for All” consortium it intends to expand, with A.X K2 at the center, into services that can find and apply for public services and even book medical appointments. The consortium includes Selectstar, which verifies data quality, and AIM Intelligence, which hunts for AI weaknesses in order to block dangerous responses and actions.
The direction SK Telecom is signaling, ultimately, is to build up its own AI capabilities while pushing safety checks outside the company’s walls. The company says experience with this kind of joint verification can become a foundation for trust with partners and customers. What to watch from here: how much joint testing actually reduces real-world risks such as information leakage or actions taken beyond granted permissions, and whether sharing verification results genuinely lowers the burden of duplicated testing.