PrivateGPT Tackles Sensitive Info in ChatGPT Prompts
With concerns among privacy advocates that employees may inadvertently enter sensitive information into the ChatGPT artificial intelligence model, the new redaction tool by Private AI has been launched. This tool aims to reduce companies' risks of exposing customer and employee data by automatically redacting 50+ types of personally identifiable information (PII) in real-time as users enter ChatGPT prompts.
PrivateGPT works seamlessly with OpenAI's chatbot, sitting in the middle of the chat process, and removing sensitive data from health data and credit card information to birthdays, Social Security numbers, and other sensitive information. The tool re-populates PII within ChatGPT answers to make the process seamless for users.
"Generative AI will only have a space within our organizations and societies if the right tools exist to make it safe to use. By sharing personal information with third-party organizations, [companies] lose control over how that data is stored and used, putting themselves at serious risk of compliance violations," said Patricia Thaine, co-founder, and CEO of Private AI.
OpenAI's LLM data set ingests every information entered into the ChatGPT prompt to train the next generation of the algorithm. Herein lies the concern of data exposure if proper security measures are not in place. A bug led to OpenAI acknowledging the leak of users' chat histories in March.
Despite OpenAI's user guide warning users to be selective when using ChatGPT, employees are still learning about privacy. Cyberhaven data security service detected and blocked requests to input sensitive data into ChatGPT from 4.2% of the 1.6 million workers at its client companies, including confidential information, client data, source code, and regulated information.
The phenomenon has caught up with Samsung engineers who made three significant leaks to ChatGPT. This included buggy source code from a semiconductor database, code to identify defects in certain Samsung devices, and the minutes of an internal meeting.
As the adoption of AI language models accelerates the delivery of code creation and analysis, "yet data leakage is most often a by-product of that speed, efficiency, and quality. Developers worldwide are anxious to use these technologies, yet guidance from engineering management has yet to be put in place on the do's and don'ts of AI usage to ensure data privacy is respected and maintained," warns Roy Akerman, co-founder, and CEO at Rezonate.
