Samsung employees disclosed confidential information to the ChatGPT chatbot at least three times, including details about semiconductors under development. The matter was reported by the Korean publication Economist.
Shortly after the tool’s release, the tech giant barred employees from using the conversational AI to protect confidential data. However, on March 11 the company lifted the restrictions, after which a leak occurred.
One employee admitted copying the entire source code of the semiconductor database loader, inputting it into ChatGPT and asking for a solution.
Another employee shared meeting notes with the chatbot and tried to automatically generate the meeting minutes. The third employee uploaded source code intended to detect faulty equipment into ChatGPT.
Thus, Samsung’s corporate secrets could be accessible to OpenAI. In the ChatGPT FAQ, говорится:
“Your conversations may be reviewed by AI trainers to improve the system.”
Developers warned that consumer services not covered by the API, such as ChatGPT or DALL-E, should be avoided to prevent personal data from entering training datasets.
Following the incidents, the tech giant implemented “emergency measures,” including a cap on upload size to 1024 bytes per prompt.
“If something like this happens even after the emergency information-security measures were adopted, access to ChatGPT on the company network may be blocked,” said Samsung leadership.
At the same time as lifting the ban on using the chatbot, the tech giant warned employees not to share confidential information with the tool.
The reports also indicate that Samsung is considering creating its own AI service to prevent similar incidents.
Impact of the leak on other companies
According to The Korea Times, the incident shook domestic tech firms, including semiconductor maker SK hynix and electronics manufacturer LG. Both firms are currently drawing up guidelines on the use of chatbots.
SK hynix reportedly blocked the use of conversational AI within its internal network. Employees who wish to use the service must obtain permission from the security department.
LG has decided to run an educational campaign to ensure its staff understand how to protect the company’s secrets.
Earlier in March, a glitch in ChatGPT disclosed some confidential data to third parties, including fragments of conversations and subscribers’ payment information.