Samsung Electronics deployed ChatGPT Enterprise and Codex to all its South Korean employees and its global Device eXperience (DX) division on June 21, 2026, reversing a company-wide ban imposed in 2023. The reversal is one of OpenAI’s largest enterprise rollouts—and its real value for operators is not the scale but the governance architecture that made it possible.

Key takeaways

• Samsung’s 2023 ban was a response to consumer ChatGPT; the 2026 deployment uses ChatGPT Enterprise, which does not train on customer data.
• Codex is now available to non-developer teams-marketing, manufacturing, corporate functions-not just engineers.
• Samsung’s semiconductor division remains under tighter restrictions: access policy can and should be granular by sensitivity level.

What Changed Between Samsung’s 2023 Ban and Its 2026 Deployment?

The 2023 incident was a governance failure, not a capability one. Within 20 days, engineers at the Device Solutions division pasted proprietary semiconductor code and meeting transcripts into the consumer version—which retains inputs for model training. Samsung banned all external generative AI tools and spent three years building Samsung Gauss internally.

ChatGPT Enterprise is a different product: it does not train on customer data, includes identity-based access management, and supports data-loss-prevention controls. The product risk that triggered the original ban has been addressed at the product tier level.

Before rolling out broadly, Samsung ran a pilot testing ChatGPT, Gemini, and Claude across 2,500 employees before selecting ChatGPT Enterprise. The Device Solutions unit, where the original leak occurred, was excluded and still operates under tighter restrictions.

The second shift is Codex availability. Originally built for developers, it is now extended to Samsung’s marketing, product, and manufacturing teams. OpenAI reports that knowledge workers now make up roughly 20% of Codex’s five-million weekly active users. Samsung is betting non-technical employees can describe what they need faster than waiting for IT.

Why Does This Matter for Operators Still Running AI Restrictions?

Many enterprise AI restrictions date from 2023-2024 and were calibrated to the consumer tools that existed at the time. The Samsung reversal surfaces one concrete question: is your current restriction based on a product capability that has since been addressed in enterprise tiers-or on a general “AI is risky” posture that has not been audited since?

Research on shadow AI adoption consistently shows that blanket bans redirect behavior rather than eliminate it. Workers move to personal devices; IT loses visibility without gaining safety. An enterprise-governed product used inside policy is safer than a consumer product used around it.

What to do now: If your organization has an AI restriction from 2023-2024, run a brief audit: Was the original concern about a specific product behavior—consumer ChatGPT retaining user inputs for training? Has the enterprise tier addressed it? If yes, request the vendor’s current data processing agreement covering data retention, training exclusions, access controls, and division-level restrictions. Before extending AI to non-developers, establish spend and token governance at the team level; automated workflows will surprise procurement teams quickly.

What to watch next: Samsung’s post-deployment metrics-productivity gains, incident counts, and non-developer Codex adoption rates-will be the first large-scale evidence of whether agentic tools work for knowledge workers in a governed enterprise environment. Watch also whether Apple, JPMorgan, and Goldman Sachs-which implemented parallel bans in 2023-revisit their policies on a similar timeline.

Frequently Asked Questions

Why is Codex being extended to non-developers at Samsung? Codex turns plain-language descriptions into working software, internal tools, and automated workflows without programming skills. Samsung is deploying it to marketing, manufacturing, and corporate teams so employees can build what they need without waiting for IT. The governance requirement: define scope and human approval gates before non-developers can build automated workflows with access to internal systems.

What is the governance difference between consumer ChatGPT and ChatGPT Enterprise? Consumer ChatGPT in 2023 could retain user inputs for model training-the mechanism behind Samsung’s data leak. ChatGPT Enterprise does not train on customer data and adds identity-based access management and data-loss-prevention controls. Operators should request the enterprise data processing agreement to confirm these terms are contractually documented, not just described in marketing.