From our correspondent in Seoul,
If ChatGPT revealed Samsung Electronics secret data, it is simply because employees of the South Korean company gave him this information… ChatGPT is a conversational agent, it is further enhanced by the exchanges it speaks with users. That is, he refines his knowledge thanks to the conversations and requests that are addressed to him. Everything that is transmitted to it is converted into learning data, making it possible to continuously improve the performance of the artificial intelligence. But the implication is that what we share with ChatGPT, we may also share with other users.
Leak in semiconductors
Company employees have apparently disclosed confidential information three times. For the first time, a person who encountered a problem running a program to download a database copied all the source code in ChatGPT so that we could offer him a solution. Then another person wanted to “tweak” some code to make it more efficient, so they grabbed all the info for the AI to do that. Finally, the third person converted the minutes of a meeting recorded on their phone and forwarded it to ChatGPT with a request for a presentation.
► To listen: [Décryptage] Chat GPT, the robot that fascinates and disturbs
Three bugs that have led to certain data leaks in the key semiconductor sector in particular. She quickly decided to limit download capacities in ChatGPT to 1,024 bytes per person and launched an investigation into the various leaks.
Samsung wants to create its own AI service
The company plans to develop its own artificial intelligence to simplify the work of its employees without revealing confidential data. Online seminars have been organized for company executives to enable them to adapt to technological developments.
The paradox is that Samsung had already taken into account that the use of artificial intelligence could generate a potential data leak, since only part of the employees were authorized to use ChatGPT to better understand its potential. Samsung seems impossible to recover the leaked data because the artificial intelligence user manual clearly stated that sensitive information should not be revealed in the conversations.