SEOUL, July 22 (AJP) - Samsung Electronics is set to deploy ChatGPT Enterprise into its prized and heavily coveted semiconductor business by the end of July, signaling a strategic shift as artificial intelligence enters one of the company's most security-sensitive operations.
The rollout marks a notable turn for Samsung's Device Solutions (DS) division, which oversees its semiconductor operations and has maintained some of the company's toughest restrictions on external generative AI since sensitive data leaks in 2023.
The chip unit had initially planned to introduce the enterprise service in June, but postponed the rollout by about a month after conducting additional security reviews involving semiconductor technology and internal data handling.
An official familiar with the matter said the internal timetable had been revised from June to July, with deployment now targeted by the end of this month.
Samsung began allowing employees across its broader corporate organization to use external generative AI services on June 12, according to a separate company official. Employees were required to complete online training covering security guidelines before gaining access.
"We made external generative AI available from June 12," the official said. "People say it is much more convenient because they can use it directly from their company PCs."
The DS division's rollout has taken longer because semiconductor engineers routinely handle highly sensitive information involving chip designs, manufacturing processes, equipment specifications, software code and production yields.
"DS has more security issues to examine than other divisions," the official said.
Samsung signed an enterprise agreement directly with OpenAI for ChatGPT Enterprise rather than deploying a separately developed in-house language model.
Unlike Samsung's earlier AI tools, ChatGPT Enterprise is expected to process much larger documents and attachments without significant performance loss, an advantage for semiconductor engineers who routinely work with lengthy technical documents, software code and extensive research and manufacturing data.
That capability could fundamentally change how engineers retrieve and use technical knowledge, according to Choi Byung-ho, a professor at Korea University's Human-Inspired AI Research Institute.
"Previously, information had to be stored in a database before it could be searched," Choi said. "Now, AI can bring together materials scattered across personal computers, cloud storage, shared folders and databases, summarize them and provide links to the original sources."
The technology could also move beyond information retrieval by using accumulated corporate knowledge to support a degree of inference, analysis and forecasting, he added.
Samsung has not disclosed what categories of semiconductor data employees will be permitted to upload into ChatGPT Enterprise. Highly sensitive information related to nationally protected technologies, chip designs, manufacturing processes, equipment and yields is expected to remain restricted under the company's internal security policies.
Nor has Samsung disclosed where employee prompts and uploaded files will be processed or stored, or detailed the architecture supporting the service.
Industry officials believe the system is likely to rely primarily on domestic servers, although the precise data-storage location and processing architecture remain unclear.
Choi said server location remained one of the most important security considerations for companies handling strategic technologies.
"To protect data sovereignty, the first question is where the server is located," Choi said. "The next question is whether the server sits inside or outside the organization and how the data is encrypted."
"Installing the servers inside the organization and cutting off external communications could be the safest option," he added. "But no technology is 100 percent secure. Ultimately, how well a company manages security is itself a measure of its capability."
The deployment represents a sharp reversal from Samsung's response to generative AI in 2023, when engineers in the chip division were found to have entered source code, equipment-related information and meeting records into ChatGPT.
Samsung subsequently restricted access over concerns that confidential information transmitted to external servers could neither be retrieved nor deleted. The DS division later required approval for many external ChatGPT uses while promoting internally developed AI tools.
As generative AI has become increasingly integrated into software development, document analysis and research, Samsung is shifting from blanket restrictions toward controlled enterprise deployment supported by internal safeguards.
The company has also strengthened employee training. Staff in its broader corporate organization were required to complete online courses covering AI security before being granted access to external generative AI services, the corporate official said.
Samsung has separately been providing AI transformation training for senior executives, including presidents and executive vice presidents. Details of training requirements and usage policies for rank-and-file employees in the DS division have yet to be announced internally.
Choi cautioned that introducing an enterprise AI platform would not automatically translate into productivity gains.
"Applying generative AI directly to business operations is not easy," he said. "A company must first process and organize its internal data, and that requires significant time and cost."
Corporate information often exists in multiple versions and formats and contains internal abbreviations, complex tables and inconsistent structures that AI systems cannot accurately interpret without additional preparation, he said.
"The information may look plausible, but companies need accurate results," Choi said. "If even one part is wrong in a highly precise task, considerable time may have to be spent reviewing the output."
Hallucinations also remain an inherent limitation because generative AI generates responses based on probabilities rather than an independent concept of truth.
"Hallucinations cannot be completely eliminated because they are part of how generative AI works," Choi said. "The important question is how much they can be reduced, and the key is to provide the system with sufficient knowledge about the company and the task."
Companies should also instruct AI systems to acknowledge uncertainty—or stop generating responses altogether—when sufficient information cannot be found, rather than producing unsupported answers, he added.
For investors, the deployment is unlikely to generate an immediate earnings impact. It nevertheless represents a significant milestone in Samsung's broader AI transformation, bringing enterprise generative AI into the daily workflow of its most strategically important business.
The semiconductor industry is characterized by long development cycles and enormous volumes of technical information. Reducing the time engineers spend searching, comparing and analyzing data could improve research productivity and accelerate decision-making over the longer term.
The corporate official said the biggest early gains were likely to come for software developers.
"It should be much better for developers, including for coding," the official said. "Being able to use it directly within the company also makes it more convenient."
Choi expects the clearest productivity improvements to emerge first in repetitive, rules-based work.
"AI can already perform well in routine tasks where the rules have been defined," he said.
"More complex work that requires discussion, judgment and decision-making remains difficult, which means people still need to give the right instructions and review the results."
Samsung is also expanding access to other external AI platforms.
Claude Code, Anthropic's coding-focused AI service, was opened in May on a priority basis for software developers rather than the broader workforce, according to the company official.
An internal notice also lists Google's Gemini for a planned November rollout, although that timetable could change following further internal reviews.
Copyright ⓒ Aju Press All rights reserved.