AI chatbots are amazing tools that can help us with various tasks, such as finding answers, creating or editing content, and more. However, they also come with some risks and limitations we must be aware of and follow when using them at work.
In this blog post, I will share with you a draft policy for using generative AI chatbots like OpenAI’s ChatGPT and Google’s Gemini that you could share with your IT or legal department as a starting point for a conversation around employee policies for AI chat. [We are not legal experts. Please have your legal department confirm & approve].
The policy draft below aims to ensure the appropriate and responsible use of these tools while protecting our organization’s assets, data, and reputation.
Whether you are a beginner or an expert exploring AI to support your work, these policy elements will help you understand the dos and don’ts of using AI chatbots at work.
Overview
While AI chatbots can perform various functions, this policy only addresses using a web-based interface to ask or “prompt” the chatbot in a conversational manner to find answers to questions or to create or edit written content.
There are, however, risks in using this technology, including uncertainty about who owns the AI-created content and security/privacy concerns with inputting proprietary company information or sensitive information about an employee, client, customer, etc., when interacting with the chatbot. Additionally, the accuracy of the content created by these technologies cannot be relied upon, as the information may be outdated, misleading, or—in some cases—fabricated.
When a user provides text into ChatGPT, the model tries to produce a ‘reasonable continuation’ of that text based on everything the model has seen — pretty much all of the internet. That means that AI chat is excellent at predicting the next word but is not an expert in truth.
Here are the additional draft elements we recommend for an employee LLM policy, but as we noted, we are not providing legal advice. Your legal counsel should review this final policy.
Purpose
With the increasing popularity of generative AI chatbots such as OpenAI’s ChatGPT and Google’s Bard, it has become necessary to outline the proper use of such tools while working at [Company Name].
While we remain committed to adopting new technologies to aid our mission when possible, we also understand the risks and limitations of generative AI chatbots and want to ensure responsible use. We aim to protect employees, clients, suppliers, customers, and the company from harm. Because ASM is dedicated to promoting and advancing microbiology sciences, using AI chat with an eye to scientific truth and accuracy is essential.
This policy aims to ensure the appropriate use of AI tools like ChatGPT to protect our organization’s assets, limit potential liabilities, and maintain the integrity, confidentiality, and availability of all proprietary and sensitive data.
Scope
This policy applies to all employees, contractors, partners, and third parties interacting with our organization’s information systems and using AI tools. It covers all relevant data, including but not limited to proprietary corporate intellectual property, trade secrets, and regulatory data.
Eligibility
This policy applies to all employees of [Company Name] and all work associated with [Company Name] that those employees perform, whether on or off company premises.
[Pick an option below]
[Option 1 – Limited Use]
Limited use of generative AI chatbots will be allowed while performing work for [Company Name] with the approval of your [manager/director/etc.]. Company email addresses, credentials, or phone numbers [can/cannot] be used to create an account with these technologies. No company data or personally identifiable information (PII) of any kind may be submitted (copied, typed, etc.) to these platforms.
Employees wishing to use generative AI chatbots must inform their [manager/director/etc.] [verbally/in writing] how the chatbot will be used. Managers must approve or deny requests within [enter number] days.
All AI-generated content must be reviewed for accuracy before relying on it for work purposes. If a reliable source cannot verify factual information generated by the chatbot, that information cannot be used for work purposes.
Acceptable uses/what AI is good at include:
- For general-knowledge questions meant to enhance your understanding of a work-related topic.
- To brainstorm ideas related to projects you are working on.
- To create formulas for Excel spreadsheets or similar programs.
- To develop or debug code to be verified before deployment.
- To draft an email or letter.
- To summarize online research or to create outlines for content projects to assist in full coverage of a topic. Only content written by employees may be included in a final product.
- [Insert additional uses here.]
- Text summarization & explanation
- Translations
- Programming
- Rewriting
- Content generation
- Prompt engineering
- Mimicking dialogue and tone
Unacceptable uses/what AI chat is not good at include:
- Using any text created by an AI chatbot in final work products.
- Copying and pasting, typing, or in any way submitting company content or data of any kind into the AI chatbot.
- Failing to properly cite an AI chatbot when used as a resource in your work product. [Include examples from your company style guide on proper citations.]
- Math
- Consistency
- Understanding context
- High-level strategy
- Reasoning and logic
- Handling uncommon scenarios
- Extended recall/memory (It can lose context within 10 minutes/during the same chat)
- To fully generate content primarily to manipulate search engines, as it’s against Google’s spam policies.
- To fully generate videos to be uploaded to YouTube without clearly labeling them as AI-generated. Creators who have not labeled their videos as such can have their channels suspended.
Any policy violation will result in disciplinary action, including termination.
[Option 2 – Open Use]
Generative AI chatbots will be allowed while working for [Company Name]. Company email addresses, credentials, or phone numbers [can/cannot] be used to create an account with these technologies. No proprietary company data or personally identifiable information (PII) may be submitted (copied, typed, etc.) to these platforms.
Employees wishing to use generative AI chatbots should discuss the parameters of their use with their [manager/director/etc.]. Managers may verbally approve, deny, or modify those parameters to meet company policy, legal requirements, or other business needs.
Acceptable uses/what AI chat is good at include:
- For general-knowledge questions meant to enhance your understanding of a work-related topic.
- To brainstorm ideas related to projects you are working on.
- To create formulas for Excel spreadsheets or similar programs.
- To develop or debug code to be verified before deployment.
- To draft an email or letter.
- To summarize online research or to create outlines for content projects to assist in full coverage of a topic. Only content written by employees may be included in a final product.
- [Insert additional uses here.]
- Text summarization & explanation
- Translations
- Programming
- Rewriting
- Content generation
- Prompt engineering
- Mimicking dialogue and tone
Unacceptable uses/what AI chat is not good at include:
- Using any text created by an AI chatbot in final work products.
- Copying and pasting, typing, or in any way submitting company content or data of any kind into the AI chatbot.
- Failing to properly cite an AI chatbot when used as a resource in your work product. [Include examples from your company style guide on proper citations.]
- Math
- Consistency
- Understanding context
- High-level strategy
- Reasoning and logic
- Handling uncommon scenarios
- Extended recall/memory (It can lose context within 10 minutes/during the same chat)
- [Insert additional uses here].
All employees using AI chatbots or other LLMs should abide by the following:
Human-in-the-loop
All AI-generated content must be reviewed for accuracy before relying on it for work purposes, and AI chat is prone to hallucination. If a reliable source cannot verify factual information generated by the chatbot, that information cannot be used for work purposes. The human review should include a Q/A content workflow that validates facts, corrects unwanted bias, and confirms accuracy and tone.
Copyright
All AI chatbot-generated content must be appropriately cited, as must the use of AI chatbot-generated content when used as a resource for company work, except for general correspondence such as email. Content from AI tools like ChatGPT is usually nonrecoverable, so it cannot be retrieved or linked in your citation and should be cited as personal communication. [Include examples from your company style guide on proper citations.] OpenAI’s policy states that “the role in formulating the content is clearly disclosed in a way that no reader could possibly miss.”
As generative AI, chatbots may produce content plagiarized from their knowledge base, including copyrighted works. No text generated or partially generated from a chatbot will be eligible for a [Company Name] copyright, trademark, or patent. AI chat images can not be granted copyright protection in the US.
[Optional: insert text explaining your company’s intellectual property policy concerning content created by or with generative AI chatbots.]
[Insert company policy regarding what % of copy should be human-generated vs created by AI, as evaluated by an AI detector tool].
Any policy violation will result in disciplinary action, including termination.
Ethical Use/Discrimination & Impact
Employees must use generative AI chatbots per all [Company Name]’s conduct and anti-discrimination policies. These technologies must not be used to create inappropriate, discriminatory, or otherwise harmful content to others or the company. Such use will result in disciplinary action, up to and including termination. When using AI chat, put yourself into the shoes of individuals or groups your creation might negatively impact.
Data used with AI
All data used with AI tools must be appropriately anonymized or pseudonymized to maintain privacy and ensure compliance with relevant laws and regulations.
- AI tools should not be used to generate, disseminate, or store any proprietary corporate intellectual property or trade secrets without appropriate safeguards.
- Data outputs from AI tools should be audited and monitored to detect and respond to any suspicious activity or potential breaches.
- AI models should be trained on data free from personal, sensitive, or confidential information to avoid unintentional leakage.
Compliance with Laws and Regulations
All activities related to the use of AI tools must comply with applicable laws and regulations, such as GDPR, CCPA, and other relevant data protection laws. The organization must fulfill all its data controller or processor obligations per these laws and regulations.
(This excellent article covers the EU AI Act, Biden’s AI order, and how it might impact your use of AI chat. The critical takeaway is that not all LLM models are compliant. Stanford Research recently published a paper titled “Do Foundation Model Providers Comply with the Draft EU AI Act?” which identifies various indicators for EU AI compliance and for your SEO work to be compliant, ideally, you use a generative AI solution that is EU AI Act-friendly or respects these criteria. Here’s how the models fared (Bloom is the most compliant).
The EU AI Act imposes fines for non-compliance, with the heftiest fines imposed for violating the prohibition of specific AI systems, up to 40,000.00 EUR or 7% turnover. The lowest penalties are for providing incorrect, incomplete, or misleading information, up to 5,000,000 EUR or 1% of annual worldwide turnover.
Training
All managers will be trained on the proper use of generative AI chatbots in the workplace.
[Optional: All employees using generative AI chatbots for work purposes must attend training on properly using these technologies before doing so.]
All questions related to this training should be addressed with [enter department and/or name and contact information].
Policy Compliance
[Title/name], or equivalent authority, will conduct regular audits to ensure compliance with this policy. They will also carry out investigations for any suspected breach or violation.
[Company Name]’s Computer Use Policy and relevant monitoring policies still apply when using generative AI chatbots with company equipment. If you have any questions regarding this policy, please contact [insert name contact information of appropriate person].
Policy Review
This policy will be reviewed and updated regularly, at least annually, to ensure its effectiveness and compliance with current legal and regulatory requirements.
Policy Updates
Different divisions and teams will be updated about ongoing AI and data initiatives via [fill in information here].
[Signatures]
Let me know if you have any feedback or have seen other clauses used in your employee policies!
Leave a Reply