Samsung Electronics' Device Solutions (DS) division is set to implement the enterprise version of the generative artificial intelligence (AI) service, ChatGPT Enterprise, by the end of this month.
Following an internal data leak incident in 2023, Samsung's semiconductor division had maintained a cautious stance on the use of external generative AI. This move marks a significant step toward integrating enterprise AI into actual operations. Given the sensitive nature of the information handled in semiconductor design, processes, equipment, source code, and yield, the division is shifting its policy from outright blocking generative AI to combining enterprise services with internal management systems while enhancing security controls.
According to a report from Aju Economy on July 22, Samsung's DS division has signed a corporate contract with OpenAI and is currently proceeding with the necessary steps to introduce ChatGPT Enterprise by the end of this month.
The initial target for implementation was set for June, but additional security checks regarding semiconductor technology and internal data caused a delay of about a month.
A Samsung official stated, "We initially considered a June rollout, but the schedule was adjusted to July. The DS division had more security considerations to review than other departments."
Since June 12, all employees across Samsung have been allowed to use external generative AI, provided they complete online security training beforehand.
However, the DS division conducted separate security reviews due to the critical information related to semiconductor design, manufacturing processes, production equipment, software code, and yield. This is why they opted for a corporate service with enhanced data protection and management features rather than a general consumer service.
The service Samsung is implementing is OpenAI's ChatGPT Enterprise, which is understood to be provided through a direct corporate contract rather than utilizing a self-developed language model.
ChatGPT Enterprise offers enhanced organizational account security and data management features compared to standard generative AI for general users. Samsung expects it to be more effective than existing AI tools in handling large documents, attachments, technical data searches, summaries, and software development support.
Previous services reportedly experienced slow processing speeds or operational difficulties when handling large volumes of data simultaneously. With the introduction of ChatGPT Enterprise, semiconductor engineers are likely to utilize it for searching, analyzing, and extracting necessary information from extensive technical documents, research materials, and software code.
The semiconductor industry is characterized by the accumulation of vast technical data across various process stages and product lines, with related information dispersed across personal computers, internal servers, shared folders, and databases. This creates a significant demand for integrated search and analysis using generative AI.
Choi Byung-ho, a professor at Korea University’s AI Research Institute, explained, "In the past, you could only search for information stored in databases, but now you can gather and summarize scattered materials from personal PCs, cloud storage, shared folders, and databases all at once, linking back to the original sources."
He added, "Generative AI can perform not just simple searches and summaries but also a certain level of inference, analysis, and prediction based on accumulated information."
However, Samsung has not disclosed the extent to which employees can input semiconductor-related data into ChatGPT Enterprise. Sensitive information designated as national core technology or related to semiconductor design, processes, equipment, and yield is likely to be restricted under internal security regulations.
Details regarding where the service will process and store data, how external and internal networks will be connected, and whether a separate system for accessing internal data will be established have also not been revealed.
The key to implementing enterprise generative AI lies in controlling how information is processed and stored after being input. For semiconductor companies, even a single piece of technical data can directly impact competitiveness, necessitating stricter requirements for server location, data encryption, access rights, and usage record management compared to general businesses.
Professor Choi emphasized, "To protect data sovereignty, the first step is to confirm where the servers are located. Whether the servers are internal or external, how the data is encrypted, and who can access it are critical security factors."
He noted, "Building servers internally and blocking external communications may be the safest approach, but there is no such thing as 100% secure technology. Ultimately, how well access rights, usage procedures, and post-incident checks are managed determines a company's capabilities."
This initiative also highlights a shift in Samsung's strategy regarding the use of generative AI.
After it was revealed in June 2023 that some employees in the DS division had input semiconductor equipment-related data, source code, and meeting notes into external ChatGPT, Samsung significantly restricted the use of external generative AI. Subsequently, a separate approval process was established for using external ChatGPT, and the company has been developing its own AI services to mitigate the risk of internal information leaks.
This incident has been cited as a notable example of the security risks associated with generative AI in the global manufacturing sector. As concerns grew that information entered for convenience could be transferred to external AI providers, Samsung prioritized information protection over generative AI utilization.
However, as generative AI rapidly becomes a key tool for software development, document creation, data analysis, and research and development, it appears that the company has recognized that a complete ban would hinder operational efficiency and technological competitiveness.
Samsung's push to implement enterprise generative AI is interpreted as a strategy to move away from a blanket prohibition on external services and instead combine enhanced security paid services with internal controls to broaden its usage.
Samsung is also expanding its AI-related training. The company is conducting AI transformation (AX) training for key executives, and all employees are required to complete online security training before using external generative AI.
To effectively apply AI tools in actual work, it is essential for employees to clearly distinguish between permissible and prohibited information and to verify the accuracy of the results.
However, there are concerns that simply implementing enterprise AI will not automatically enhance productivity. Without organizing internal data in a format that AI can utilize and removing outdated or duplicate materials, it will be challenging to expect accurate responses.
Professor Choi stated, "To apply generative AI in actual work, the process of adequately organizing and processing internal data must occur first. Internal company materials vary in format and version, and even the same information can differ based on the time of writing and department, making it difficult for AI to understand accurately."
He added, "What companies want is not plausible answers but accurate ones. If even one error occurs, it can take considerable time and manpower to verify the results again."
The issue of 'hallucination,' where generative AI presents false information as if it were factual, remains a challenge. In semiconductor design or production environments, using incorrect information can impact not just document accuracy but also development schedules, productivity, and quality control.
Professor Choi noted, "Hallucination arises from the structure of generative AI generating responses based on probabilities, making it difficult to eliminate entirely. It is crucial to provide AI with sufficient knowledge about the company and work, and to ensure it responds only within reliable data to minimize errors."
He further stated, "If AI cannot find sufficient information, it should be programmed to respond with 'I don't know' or to halt work. A system that presents the original text and sources that underpin the answers is also necessary."
Samsung is gradually introducing other generative AI services alongside ChatGPT. The coding support service 'Claude Code' was opened to software development personnel in May.
Internal notices from Samsung indicate plans to introduce Google's generative AI service 'Gemini' in November. However, this schedule may change based on internal security reviews and system readiness.
Industry observers are paying attention to the possibility that Samsung will establish a system to selectively utilize multiple generative AIs, such as ChatGPT Enterprise, Claude Code, and Gemini, based on the characteristics of different tasks. Since each service has different strengths for document search and analysis, software development, and research and development, it may be more effective to apply services according to their intended use rather than relying on a single AI.
The introduction of ChatGPT Enterprise in Samsung's DS division is expected to be a test of finding a balance between utilizing generative AI and ensuring semiconductor technology security. For this initiative to lead to actual productivity improvements, it is crucial to establish sophisticated operational standards that encompass the scope of input data, access rights, response verification, and accountability in the event of incidents.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.