I am often asked by colleagues, “Do you have a sample ChatGPT policy for employers?” Although it would be convenient for all involved if there was a “one size fits all” policy on this issue, there is none, nor can there be one. Each employer—based on its workplace culture, business model and employee base—needs to decide, after considering a number of factors, whether and how to restrict the use of ChatGPT in the workplace. ChatGPT is a form of artificial intelligence (AI), or more specifically, an AI language learning model created by OpenAI that has been trained through consumption of large swaths of data from the Internet and interacts in a conversational way when responding to queries or requests from users.
June 02, 2023
ChatGPT in the Workplace: To Restrict or Embrace, That Is The Question
Karla Grossenbacher
There are two primary legal risks for employers to consider when thinking about how ChatGPT might be used in the workplace: bias and confidentiality. ChatGPT’s ability to provide responses to users is dependent on and limited by the data upon which it was trained. Bias can be generated in AI through the information on which the AI was trained and the decisions made by the human(s) who decided on which information to train the AI. If employees use ChatGPT in connection with employment decisions, this could potentially lead to legal liability for the employer. For example, employees could ask ChatGPT to evaluate the qualifications of candidates or the written work product of employees. In addition to the use in employment-related decision-making, ChatGPT could be used to perform certain tasks that do not constitute decision-making with respect to employment, but could bear on it. For example, an employee could ask ChatGPT to create the content for a job description and identify the essential duties or experience needed for a given position, and bias could creep in. There are also laws in a number of jurisdictions that place restrictions on an employer’s ability to use AI in hiring and using ChatGPT in connection with employment decisions can create related compliance issues.
The bigger and more obvious concern with allowing employees to use ChatGPT in connection with their jobs is its potential impact on the confidentiality and data privacy of Company information. As a language learning model, ChatGPT does have the ability to “learn” from every conversation. ChatGPT was trained on large amounts of data from the Internet, but conversations with ChatGPT can be used to provide additional training to fine-tune ChatGPT. Currently when a user logs into ChatGPT, a screen pops up that states “[c]onversations may be reviewed by our AI trainers to improve our systems” and advises users “[p]lease don’t share any sensitive information in your conversations.” Thus, if employees are entering confidential employer information into ChatGPT, it is clear that such information will be used and shared with OpenAI to a certain degree. Accordingly, a compelling argument could be made that an employer’s failure to prohibit employees from inputting confidential and proprietary information into ChatGPT is not consistent with treating such information as a “trade secret,” and the information at issue could lose its status as a trade secret due to the employer’s failure to protect it. Another related issue is that employees who use ChatGPT for their duties could unwittingly be exposed to the confidential information of others and use it in connection with their work. ChatGPT was trained on wide swaths of information from the Internet, and it is conceivable that ChatGPT absorbed information that, while publicly available on the Internet, was trademarked, copyrighted and/or the intellectual property of another person or entity. If an employer’s employees take possession of or use such information, this could create additional legal risk for the employer. As the foregoing demonstrates, there are a number of risks and potential benefits that come with using ChatGPT in the workplace. Employers who have not addressed or grappled with these issues will need to do so in the coming months as other AI language learning models have entered, and will continue to enter, the market. This will require a deft balance against potentially leveraging such AI to create efficiencies in how employees work while protecting employer confidential information and protecting against workplace bias.