SK Telecom to test AI model safety at MWC26

SK Telecom will join the “Global AI Red Team Challenge” at MWC26 in Barcelona on March 3, aiming to assess the safety of its AI model A.X K1.

The event involves evaluators testing AI systems from various telecom and AI firms for issues like bias or data breaches. Co-hosted by GSMA and SelectStar, the challenge does not publicly disclose company rankings.

SK Telecom’s A.X K1 is an internally developed large language model selected for a national AI project. The company intends to use the event to verify its model’s safety and improve its AI services. Kim Tae-yoon, head of SK Telecom’s foundation model division, said safety must go hand in hand with AI development.

🔗 Source: Chosun Daily

🧠 Food for thought

Implications, context, and why it matters.

SKT’s model is part of South Korea’s sovereign AI strategy

A.X K1 is a 519-billion-parameter system built under South Korea’s government-led “Sovereign AI Foundation Model” project 1.
The project builds a domestic AI value chain, from chips to services, to cut reliance on foreign technology. South Korea also targets a top-three global spot in AI 1.
SKT calls the system a “Teacher Model” and labels it “digital social overhead capital” (SOC). The goal involves shared digital infrastructure that passes know-how to smaller, specialized AI systems across Korean industry 2.
SelectStar co-hosted the event and also belongs to the SKT consortium that built A.X K1. SelectStar handled data construction and validation work, rather than acting only as an independent evaluator 1.

Sovereign AI models could create a new class of full-stack competitors

SKT’s public safety test at the “Global AI Red Team Challenge” fits a wider telecom push beyond connectivity into direct competition with cloud hyperscalers (the biggest cloud-computing providers) 3.
With AI-specific data centers plus foundation models, SKT can sell a “full-stack” sovereign option. This setup keeps sensitive data inside national borders for government and enterprise clients 1.
The same build-out can support a domestic semiconductor industry. The hyperscale model checks the performance of AI chips and memory 2.
Many countries now build AI platforms rooted in local data and culture. Gartner (a technology research and advisory firm) expects 35% of countries will adopt region-specific AI platforms using proprietary contextual data by 2027 1.