Artificial Intelligence | June 21, 2023
Google continues to refine the responses of its chatbot, Bard. However, while it may respond well, it is not recommended from a security standpoint. In fact, the company advises its employees to exercise caution when using such language models.
Internal sources at Alphabet have explained to Reuters that the company has recommended its employees not to input any kind of confidential information into chatbots, be it the popular ChatGPT, GPT-4, or even their own chatbot, due to the risk of information leaks. It appears that Alphabet is concerned about the possibility of its employees entering sensitive or confidential data into these AI tools. So far, Google has not responded to Gizmodo’s request for further comment on this matter.
The reason for this is quite evident, as most chatbots have human reviewers behind them who check the data we input to help refine the language model’s responses. The presence of these technicians can lead to leaks, and this is not the first time it has happened. Last month, Samsung confirmed that it had experienced leaks of confidential data after its staff used ChatGPT.
In January of this year, an Amazon attorney urged their staff not to share code with ChatGPT. Slack screenshots revealed by Insider showed that the attorney requested employees not to share confidential information with the chatbot, nor any programming code they were working on.
Last month, Apple also sent a similar notice to its employees, as revealed by internal documents leaked to The Wall Street Journal. Cupertino went a step further and explicitly prohibited the use of both ChatGPT and Microsoft’s Copilot tool, which facilitates coding on GitHub. The same leak revealed that Apple is interested in developing its own language model, which is why they have already acquired two young AI companies for $200 million and $50 million.
Google launched Bard in March of this year to compete with ChatGPT. Alphabet’s chatbot is based on its own language model called LaMDA (Language Model for Dialogue Applications). Just before that, an internal memo revealed that Google CEO Sundar Pichai asked employees to test Bard for two to four hours a day. The chatbot has been delayed in the EU due to allegations from Ireland that it does not comply with European data protection laws.
