Identifying data
Samsung: source code and a meeting transcript in ChatGPT
In March 2023, within 20 days, three employees in Samsung's semiconductor division gave ChatGPT confidential material: source code twice and a meeting transcript once. The company later restricted generative AI.
Samsung Electronics' semiconductor division allowed staff to use ChatGPT from 11 March 2023, while its mobile and consumer electronics division kept its ban. According to a report by the Korean edition of the Economist on 30 March, fewer than 20 days later the company had learned of three incidents.
One employee pasted in the entire source code of a program that downloads data from a measurement database for chipmaking equipment, to find a bug in it. Another entered the code of a program written to track yield and faulty equipment, and asked for it to be optimised. The third recorded a meeting on a smartphone, had the recording turned into text with the Naver Clova app, and then gave that text to ChatGPT to produce minutes.
Samsung limited uploads to 1,024 bytes per question, launched an investigation and said staff could face disciplinary action where necessary. It warned employees that anything entered into ChatGPT goes to external servers and cannot be retrieved by the company. It also considered building its own in-house AI service. The company did not respond to The Register's request to confirm the details.
In early May, Bloomberg reported on an internal memo: Samsung had temporarily banned generative AI tools such as ChatGPT, Google Bard and Bing on company computers, tablets and phones and on its internal networks. In an internal survey in April, 65% of respondents saw these services as a security risk. According to the company, data sent to such platforms is hard to retrieve and delete and could be disclosed to other users; breaking the rules could lead to dismissal.
What can we learn from it?
The third incident is the most telling: a recorded meeting went first to a transcription app and then to a chatbot, just to produce minutes. The goal itself is entirely sensible; AI is excellent at summaries and minutes. The difference lies in the route: where the transcript is made, and who receives the text. In Confiella the recording never leaves the computer, the transcript is made locally, and only reviewed, pseudonymised text goes to the cloud. Whatever tool we use, it is worth asking those two questions at every step.
Sources
- Economist Korea: [단독] 우려가 현실로…삼성전자, 챗GPT 빗장 풀자마자 '오남용' 속출
- The National: Samsung Electronics bans employees from using ChatGPT
- The Register: Samsung reportedly leaked its own secrets through ChatGPT
This article is based on the public sources listed. If you find a mistake, write to us: [email protected]