Samsung's 2023 ChatGPT incident is still one of the most referenced data-security stories in corporate history. Within just 20 days of letting engineers use the free public version of ChatGPT, three separate leaks occurred. Engineers pasted semiconductor source code, equipment defect algorithms, and transcripts from internal meetings directly into the tool. Once submitted, Samsung had no way to retrieve or delete that data.
The company responded by banning ChatGPT and similar tools entirely on company devices and networks. It also began building its own in-house AI model, called Samsung Gauss, to keep sensitive work within controlled infrastructure.
Fast-forward three years. Samsung ran a structured two-month trial earlier in 2026, with 2,500 employees testing enterprise versions of ChatGPT, Google's Gemini, and Anthropic's Claude. That pilot led directly to the full deployment announced on June 21, 2026. OpenAI describes it as one of its largest enterprise deals ever. The rollout covers all Samsung Electronics employees in South Korea and all employees globally in its Device eXperience division, which handles smartphones, consumer electronics, and home appliances.
What changed is not Samsung's risk appetite. What changed is the product. The enterprise version of ChatGPT does not use customer data to train its models by default. Company administrators control who has access, can see a full audit trail of usage, and can revoke access at any time. Samsung also built a requirement into the rollout: employees only get access after completing internal security training.
The tools are not limited to technical staff. ChatGPT will be used for document drafting, information search, idea development, and data analysis across marketing, manufacturing, and corporate functions. Codex, which helps people write and review software code, is also being rolled out to non-technical teams so they can build small internal tools and automated workflows without needing a software engineer.
This deployment sits inside a much larger relationship. Samsung and OpenAI signed a broader strategic partnership in October 2025 covering AI infrastructure. Samsung's chip division supplies memory components for OpenAI's data centres. Samsung SDS, the company's IT services arm, has become the first Korean company authorised to resell ChatGPT Enterprise to other Korean businesses.
The adoption numbers from Korea suggest this was already happening informally before the formal deal. Weekly active users of OpenAI's Codex tool in South Korea grew by nearly 800% between February and June 2026. That kind of growth, before a formal company-wide rollout, is a reliable sign that employees were finding the tools and using them regardless of official policy.
That is the practical point for any organisation still sitting on a blanket AI ban. Banning the free public version does not stop usage. It shifts it to personal devices, personal accounts, and zero oversight. The enterprise versions with proper access controls and data protections are structurally different products from the free consumer tools that caused Samsung's original problem.
The broader shift is visible across the market. A Deloitte survey from 2026 found that 66% of organisations reported productivity gains from enterprise AI adoption. At the same time, a separate survey of French mid-sized companies found that only 17% of those using AI reported actual time savings, which suggests deployment alone does not guarantee results. Samsung's approach, piloting with 2,500 employees first and requiring security training before access, is closer to how productive deployments actually work than simply switching the tools on and hoping for the best.