On Monday June 22, 2026, OpenAI announced that ChatGPT Enterprise and Codex will be deployed to every Samsung Electronics employee in Korea and every worker in its Device eXperience division worldwide. The deal is one of OpenAI's largest enterprise rollouts to date. Samsung's DX division — visual displays, digital appliances, health and medical equipment, mobile, and networks — accounts for sixty-two percent of company sales and is the customer-facing half of an employer with more than 267,000 staff globally. Three years and three months ago, in March 2023, Samsung banned ChatGPT and every other generative AI tool company-wide after its own engineers pasted sensitive source code and internal meeting notes into the chatbot. The ban became the canonical case study for what enterprises were not allowed to do with the technology. The rollout, today, is the canonical example of what enterprises now do with it.

The reversal, in two dates

Samsung's March 2023 prohibition was quoted in every enterprise-AI risk presentation written that year. The trigger was three separate incidents in twenty days where Samsung Semiconductor employees pasted confidential material into ChatGPT — a chunk of proprietary source code, test-pattern data for chip yields, and a recording of an internal meeting transcribed by a third-party app. The reaction was a blanket ban on any generative tool on any company device for any work purpose, with dismissal threatened for further violations.

The June 22, 2026 rollout covers chip design engineers, customer service teams, R&D, manufacturing, marketing, and corporate functions. The two memos describe the same company three years apart, the same vendor on the other side of the same conversation, and the only difference is that Samsung now thinks it has figured out how to do this safely. That conclusion is the thing the rest of the Fortune 500 has been waiting for. They now have a reference customer.

The proof of concept that did the work

The detail in the Korea Times reporting and the Asia Business Daily writeup that does not show up in the OpenAI press release is the security architecture Samsung built to get from the 2023 memo to the 2026 rollout. From April to May 2026, the DX division ran a two-month proof of concept with 2,500 employees evaluating ChatGPT, Gemini, and Claude side-by-side against internal workflows. The output of the POC was not a model choice — Samsung is keeping all three — but a control system. Tool access is gated on completion of an internal security training programme, the kind that auto-revokes on policy lapse, and the deployment runs through Samsung SDS, the company's IT services arm, which is now the first Korean entity authorised to resell ChatGPT Enterprise to other businesses. The reseller agreement is the second story buried inside the first one. Samsung is not just a customer. It is a channel.

Roh Tae-moon, the President and Head of Samsung's DX division, described the shift as structural rather than incremental. The phrasing matters. Companies that frame an AI deployment as a productivity upgrade are talking about the next quarter. Companies that frame it as structural are telling investors that the operating model has changed. Roughly five million people now use Codex every week, the OpenAI announcement says — and Codex weekly active users in Korea grew nearly eight hundred percent between February 1 and the rollout date, four and a half months of installed-base growth that arrives mostly through one customer's onboarding.

What it means that ChatGPT is not the only one

Samsung is not buying exclusivity. The same Korea Times piece names Claude and Gemini as live alternatives inside Samsung, and the POC tested all three. Samsung sells its phones, displays, and televisions to customers who use every model in the market — Galaxy AI on the device side already brokers Google's Gemini for some queries and Anthropic's Claude for others, depending on the task. The fact that ChatGPT Enterprise sits alongside Gemini and Claude rather than in place of them is the part that distinguishes this from the kind of single-vendor lock-ins SAP and Oracle wrote in the 1990s. The enterprise procurement model that is emerging — at Samsung, at JPMorgan, at the firms that signed OpenAI's Partner Network eight days ago — is a multi-LLM stack with a control plane on top. Samsung SDS is the control plane.

That is the part the OpenAI announcement gets to claim and not claim at the same time. One of our largest enterprise deployments, yes. Exclusive Samsung AI vendor, no. The press release is careful to say the former and silent on the latter.

The Stargate triangle closes

Eight months before the rollout, on October 1, 2025, Sam Altman flew to Seoul for meetings at the presidential office with President Lee Jae-myung, Samsung Executive Chairman Lee Jae-yong, and SK Group Chair Chey Tae-won. The output was a set of letters of intent committing Samsung and SK Hynix to scale high-bandwidth DRAM production for OpenAI's Stargate programme — up to nine hundred thousand wafers per month, a volume TechCrunch reported as roughly forty percent of current global DRAM output. Samsung C&T, Samsung Heavy Industries, and Samsung SDS signed a parallel agreement to assess data centre capacity in Korea. The October memos made Samsung an OpenAI supplier. The June 22 announcement makes Samsung an OpenAI customer. Samsung's chip teams will now design the HBM that runs the data centres that host the models that Samsung's marketing teams will use to write campaign briefs. The flagship customer that supplies the silicon is not a phrase that has been writable about an OpenAI contract before. It is now.

What this is

The Loop has written, over the last fortnight, about the model wars becoming the channel wars and about the agent web becoming an open standard. The Samsung rollout is the third leg of the same shift. The model layer is no longer where the customer chooses. The customer chooses the implementation harness — and once the harness is in place, the model behind it can be swapped without renegotiation. Samsung's two-month POC and Samsung SDS's reseller licence are the harness. OpenAI got named as one of three vendors plugged into it.

The 2023 ban was the canonical worked example of an enterprise refusing to take the risk. Three years on, the same company has produced the canonical worked example of one taking it — by writing the security architecture, training the workforce, and incorporating the reseller licence as a revenue line of its own. The shift the larger market needed was not better models. It was the assurance that the failure mode of March 2023 had a repeatable fix. Samsung has now shipped the fix at full scale on its own employees.

What to watch

1. The Samsung SDS reseller pipeline. Becoming the first Korean entity authorised to resell ChatGPT Enterprise is a margin play that pays for the internal rollout. The number to watch is how many of Samsung SDS's existing IT-services customers — LG, Hyundai, the Korean banks — pick up an SDS-mediated ChatGPT licence inside Q3. Five customers means SDS is the channel. Twenty means the Korean enterprise market is consolidating onto one reseller.

2. The DS division. Samsung's chip-making half — the Device Solutions division under Vice Chairman Jun Young-hyun — is conspicuously absent from the rollout. DS is where the March 2023 leak originated and where the most sensitive intellectual property lives. If the company extends the deployment to DS inside the next two quarters, it is signalling that the control architecture holds against the exact failure mode that produced the ban. If DS stays out, the rollout's enterprise-reference value drops by half.

3. Anthropic and Google's Korean response. Samsung is running Claude and Gemini alongside ChatGPT. The vendors on the losing end of an enterprise-flagship narrative tend to publish a counter-flagship inside ninety days. Watch for an Anthropic Korea or Google Cloud Korea announcement naming a Korean chaebol — LG, Hyundai, Lotte — by the end of August. If it does not come, Samsung's tri-vendor strategy is in danger of consolidating onto whichever vendor moves first to embed its training programme into Samsung SDS's reseller catalogue.

4. The cost line. Well over a hundred thousand Samsung employees on ChatGPT Enterprise at the standard sixty-dollars-per-seat enterprise list price is a nine-figure annual line item before any custom Codex provisioning. The number Samsung is actually paying is unpublished and probably half list on volume — but the visible ceiling is large enough to show up in OpenAI's S-1 narrative as a named anchor customer. That is a sentence that costs a hundred million dollars. Whether OpenAI lets analysts say it out loud in the next earnings cycle is the cleanest read on how confident the company is that the deal locks in.

Three years ago the canonical risk story was an engineer pasting source code into a chatbot. Today the canonical risk story is a Fortune 500 board signing off on every employee using one. The intervening work was not a better model. It was the security architecture, the procurement playbook, and the partner-firm reseller licence — exactly the layers the labs spent the last six months building. Samsung is the customer who proves they shipped them.