Samsung Electronics is expanding ChatGPT Enterprise and Codex access to all employees in Korea and the Device eXperience division worldwide. The company expects the tools to support software, marketing, manufacturing, product development and everyday knowledge work.
The rollout is notable because Samsung restricted public generative-AI tools in 2023 after employees uploaded sensitive information. Its return shows how enterprise controls can change the answer from prohibition to managed use—but only if governance reaches daily behaviour.
Enterprise access replaces unmanaged experimentation
ChatGPT Enterprise gives Samsung central user administration, security controls and data protections unavailable when staff use personal accounts. Employees can search information, analyse material, draft documents and develop ideas inside an approved environment.
Codex adds code writing, review and debugging, as well as support for prototypes, websites and automated internal workflows. OpenAI says the tool has more than five million weekly users, with Korean usage growing sharply during 2026.
Broad access reduces shadow AI by giving employees a sanctioned option. It does not make every input appropriate. Product roadmaps, unreleased designs, customer information and semiconductor process data need classification rules that determine which tools and connectors may receive them.
Training should use realistic Samsung examples rather than a generic warning not to share secrets. Employees need to recognise sensitive context hidden inside logs, screenshots and long documents.
Administrators should monitor patterns without reading every employee conversation. Aggregate signals can reveal repeated policy violations, weak training or an overloaded use case, while targeted investigation remains tied to a documented security reason. Clear notice about monitoring helps prevent the enterprise environment from feeling like covert workplace surveillance.
Codex can spread software creation beyond developers
Non-technical teams may use Codex to build small tools and automate repetitive work. That lowers the barrier to solving local problems, but it can also create unsupported applications outside normal engineering oversight.
Samsung needs a path from personal prototype to maintained system. Code affecting data, customers or production should enter version control, testing, security review and named ownership before wider use.
Technical teams should measure defects, review effort and maintainability alongside time saved. Generated volume is not productivity when experienced engineers spend the gain untangling unfamiliar code.
Manufacturing use deserves particular caution. An AI suggestion can inform analysis or draft a procedure, while changes to equipment and quality systems remain protected by deterministic controls and qualified approval.
The strongest outcome is not that every employee becomes a programmer. It is that people can express workflow needs more directly while specialists preserve architecture, security and reliability.
The partnership reaches from employee tools to infrastructure
Samsung and OpenAI already have wider ties. Samsung companies are exploring data-centre services and supplying advanced memory for Stargate, with projected demand reaching hundreds of thousands of DRAM wafers per month.
That relationship places Samsung on both sides of the AI economy: its staff consume models while its semiconductor and services businesses help build the infrastructure that runs them.
Vendor concentration should still be managed. Workflows, prompts and internal connectors need documentation and portability in case pricing, capability or policy changes.
Adoption metrics should focus on completed work, decision quality and avoided rework. External surveys show many companies report broad productivity gains, yet far fewer can identify concrete time savings.
Samsung’s earlier restriction was a rational response to uncontrolled use. The new deployment will succeed if it preserves that security lesson while enabling useful experimentation.
Enterprise AI becomes sustainable when employees know where the boundary lies, managers can see how systems are used and a fast prototype does not bypass the controls protecting a global manufacturer. Reopening access is the beginning of that operating model, not proof that the risk has disappeared.
