News · 2026-06-24
Samsung Banned ChatGPT in 2023. Now It's Giving It to 125,000 Workers.
In 2023, Samsung became the textbook example of corporate caution about AI. After engineers accidentally pasted sensitive internal information into ChatGPT, the company banned the tool outright -- a story that got cited for years as proof that serious enterprises could not trust public AI with their secrets. This week Samsung completed the about-face. According to reporting from PYMNTS, the company has rolled OpenAI's enterprise products -- ChatGPT Enterprise and the Codex coding tool -- out to roughly 125,000 employees in South Korea, plus staff in its global device division, in one of the largest enterprise deployments OpenAI has ever announced.
The reversal is the story. A few years ago, the conventional wisdom in big, security-conscious companies was defensive: keep public AI tools at arm's length until the data risks are understood. The worry was concrete and reasonable -- if your employees feed confidential designs or source code into a chatbot, where does that information go, and could it leak or train a model a competitor also uses? Samsung's 2023 ban was the most famous expression of that fear. The deployment this week is a signal that the fear has been outweighed, at scale, by the productivity case -- and that the enterprise versions of these tools, which come with contractual promises that company data is walled off and not used for training, have done enough to satisfy a company that got burned badly enough to ban them once.
The scope is what makes it notable. This is not a cautious pilot with a hand-picked team. Samsung is putting these tools across software engineering, product development, marketing, and even manufacturing -- treating AI not as a specialist gadget for a few departments but, in OpenAI's framing, as a core platform for how the whole workforce operates. There is also a neat piece of mutual dependence underneath it: Samsung is one of the suppliers of the advanced memory chips that OpenAI's own AI infrastructure runs on. The customer relationship runs in both directions.
The analogy is a bank that once forbade employees from using their phones at their desks, then a few years later hands everyone a company smartphone and builds its workflow around it. The reversal is not a sign the original worry was foolish -- it was sensible for its moment. It is a sign that the technology matured, the guardrails got built, and the cost of staying on the sidelines came to outweigh the risk of joining in. That is the throughline connecting this to other reversals landing the same week, including a major stock-image company settling into partnership with OpenAI after suing one of its rivals over AI training just a couple of years ago. The pattern is consistent: the loudest holdouts are not just relenting, they are signing up, on terms they negotiated.
Why this matters: enterprise adoption is where AI either becomes a durable business or stays a consumer novelty, and the conversions of the most prominent skeptics are the clearest evidence of which way it is going. When the company that wrote the cautionary tale becomes a flagship customer, it tells every cautious competitor that the safe-by-default posture is no longer obviously the safe choice -- that the bigger risk may now be falling behind. It also raises the stakes on every concern in this week's news, because the more deeply a workforce of 125,000 leans on an outside provider's tools, the more it matters that those tools stay affordable, stay available, and do not vanish on a government order the way a frontier model just did.
The honest caveat is to read the announcement for what it is. 'Rolled out to 125,000 employees' is a measure of access granted, not of value delivered -- handing every worker a powerful tool is the easy part, and the history of enterprise software is full of expensive deployments that employees barely touched. Whether Samsung's people actually use these AI agents for work that matters, whether the productivity shows up in results rather than press releases, and whether the data guarantees hold up over years are all open questions that a launch-day headline cannot answer. The reversal is real and meaningful as a signal of where corporate sentiment has landed. The return on it is something only the next few years of actual usage will reveal.