Samsung Employees Accidentally Leaked Company Secrets Via ChatGPT: Check What Happened Next

File Photo

New Delhi: Along with the launch, ChatGPT expanding its space in the tech sector. Many IT firms, notably Samsung, have let employees use ChatGPT while carrying out their jobs. Unfortunately, this got workers into trouble when they unintentionally gave the AI chatbot access to private information. The semiconductor division of Samsung had given permission for its engineers to use ChatGPT to assist in resolving issues with source code.

Top-secret information, including the source code for a new software and meeting notes about their hardware, were accidentally input by the staff. Three such events were reported in just one month. ()

Significantly, ChatGPT keeps the data it receives to continue learning. Thus, the developer of AI chatbots OpenAI now has access to these Samsung trade secrets. ()

According to reports, Samsung Semiconductor is developing its own AI for usage by staff members inside to prevent such errors from occurring in the future. Unfortunately, it will only be capable of handling prompts that are less than 1024 bytes in size.

In one of the mishaps, a worker was optimising test sequences for secretive company chip flaw detection using an AI chatbot. A semiconductor database download program's source code was copied by the employee, who then entered it into ChatGPT and checked for problems.

Another instance was a worker using ChatGPT to turn meeting notes into a presentation. These notes contained private information that should not have been disclosed to outside parties.

Significantly, the ChatGPT FAQ for OpenAI states unequivocally that "Our AI trainers may review your interactions to improve our systems."

According to a report in The Register, Samsung's CEO has advised staff members to avoid such errors. Access to ChatGPT may be restricted on the corporate network if a similar incident takes place despite emergency information protection procedures being put in place, he has warned.

Particularly noteworthy is the statement made by OpenAI that "We delete any personally identifying information from data we intend to use to improve model performance. In addition, for our attempts to enhance model performance, we only use a tiny sampling of data per customer.