Samsung deploys ChatGPT Enterprise and Codex to global workforce, reversing 2023 AI ban
Samsung Electronics announced on June 21-22, 2026 that it will deploy ChatGPT Enterprise and Codex to all employees in Korea and to all Device eXperience (DX) division employees worldwide. OpenAI describes this as one of its largest enterprise deployments ever. The move marks a dramatic reversal: in March 2023, Samsung had banned generative AI tools company-wide after engineers accidentally leaked sensitive source code through ChatGPT. Now, with enterprise-grade security controls mandated for all users, the company is embracing AI as a core platform for productivity across R&D, manufacturing, marketing, and corporate functions.
The scale of adoption is notable. Codex weekly active users in Korea surged nearly 800% between February 1 and June 2026, before the full deployment was even announced. Samsung ran a two-month proof-of-concept in its DX division from April to May with 2,500 employees testing ChatGPT, Gemini, and Claude, then built access controls requiring internal security training. This time, none of employee-generated data, prompts, or code can be used to train OpenAI's models; all data is sandboxed in a strictly managed corporate cloud.
Beyond the internal adoption, Samsung SDS (the group's IT services arm) became the first Korean entity authorized to resell ChatGPT Enterprise and manage deployments for other businesses. OpenAI's relationship with Samsung expands from hardware (HBM semiconductors for AI infrastructure) into the enterprise software layer, making Samsung a regional technology partner in Asia-Pacific.
For executives tracking enterprise AI vendor lock-in and multi-vendor strategies, this signals that large manufacturers are moving past pilots: Samsung tested OpenAI, Google, and Anthropic before deploying ChatGPT as its primary tool, suggesting that by mid-2026, internal capability comparison is routine for major enterprises. The 800% surge in Codex usage pre-announcement also hints that broad employee adoption is achievable when tools are framed as productivity infrastructure rather than experimental features.