Apple, a company renowned for its commitment to privacy, is apparently concerned about the potential for its employees to inadvertently disclose confidential information when using ChatGPT, an AI tool developed by OpenAI. Apple has reportedly placed restrictions on the use of ChatGPT and other AI technologies such as GitHub’s Copilot, which helps autocomplete code, to address this issue. Apple is also reportedly working on creating its own substantial language models.
Three Samsung employees allegedly gave ChatGPT access to private information earlier this year. One employee asked for help finding bugs in the database source code, while another asked for code optimisation. The third employee gave the chatbot a recording of a meeting and instructed it to take minutes. Samsung restricted its employees’ use of ChatGPT by limiting prompts to one kilobyte or 1024 characters, while Apple’s restrictions on generative AI tools remain unknown.
As users engage with large language models such as ChatGPT, the models improve as the data is fed back to the creators for further training. OpenAI’s terms and conditions state that AI trainers can examine discussions to improve the system. Apple’s move to restrict the use of such technologies is not surprising, given its penchant for secrecy. In April, in response to privacy concerns, OpenAI introduced a privacy control feature that allows users to turn off chat history storage and prevent their chats from being recorded for training purposes. The feature was added after a bug in ChatGPT allowed users to see other people’s conversation history, forcing OpenAI to briefly shut down the service.
There are few, if any, details available about Apple’s own major language modelling initiatives. However, John Giannandrea, who previously ran Google’s search and AI teams, is in charge of the company’s AI ambitions. Apple has also recently bought a number of AI companies, underlining its commitment to making progress in the field. Tim Cook, Apple’s CEO, stressed the importance of a careful and methodical approach to such technologies when asked about AI in a recent interview.