As you embark on the adoption of AI in your instructional and/or research activities including AI testing, proof of concepts, and similar endeavors, it is important that members of our community are well-informed about the best practices to ensure the secure and responsible use of these powerful tools. Generative AI and other forms of AI can be incredibly beneficial for research, writing and other academic tasks, but they also come with potential risks. To protect our university's data and reputation, please follow the best practices below when using AI tools. You can also refer to Bentley University IT's Acceptable Use Policy and Guidelines for Responsible Use of AI (both are available here).
Thank you for your attention to these best practices and suggestions. Let's continue to use these powerful tools responsibly and make valuable contributions to our academic community.
Data Protection & Secure Use
Protect Bentley, student, and personal information when using AI. Avoid sharing sensitive or protected data in public AI tools, use Bentley-authorized solutions for institutional information, and verify that you are signed into a university-protected environment before working with Bentley data.
Responsible & Ethical AI Use
Use AI thoughtfully and remain responsible for the work you create with it. Review AI-generated content for accuracy and bias, be transparent about AI assistance when appropriate, and consider copyright, licensing, and intellectual property before using or sharing AI-generated materials.
Purchasing & Adopting AI Tools
Connect with ATC and Bentley IT before purchasing or adopting a new generative AI tool. We can help evaluate your use case and ensure that appropriate security, privacy, and data protection considerations are addressed before the tool is implemented.
Building & Using AI Agents
Bentley is developing formal guidance for Copilot agents. In the meantime, faculty should limit use to low-risk academic work, avoid sensitive or confidential data, verify outputs, maintain human oversight, and consult ATC with questions.
Best practices to secure university data when using AI
Only use generative AI models provided by reputable, secure, and Bentley-authorized platforms. For now, Benley’s authorized platform is Microsoft Copilot. If you log into Microsoft’s Copilot with your Bentley account credentials, data you enter or extract from the model using your chats will not be shared with others. Microsoft does not use your data to train their models and stores any data that you share securely in Bentley’s environment.
Keep your Bentley University credentials secure and private. Do not share your account details with others, and avoid using the same password for multiple accounts.
If you encounter any suspicious or unusual activity while using ChatGPT or other chatbots, immediately report it to helpdesk@bentley.edu.
Some Suggestions for Reducing Generative AI Hallucinations
Generative AI models are trained with vast quantities of internet data, which can sometimes result in responses containing hallucinations and inaccuracies. Because of this, users should critically review and verify all generative AI output. Users can also utilize more than one generative AI as an information resource.
Some additional strategies that generative AI creators/developers can consider for mitigating hallucinations and inaccuracies include prompt engineering and Retrieval-Augmented Generation (RAG). Prompt engineering techniques include providing the generative AI model with system messages to guide its overall tone and purpose and examples (question-answer pairs) that it can mimic in its responses.
University Policies & Support
AI use should remain consistent with Bentley University's existing policies and requirements, including those related to Acceptable Use Policy, Data Classification Policy, Digital Millennium Copyright Act (DMCA) Policy, Information Security Policy and Academic Integrity Policy.
Faculty, staff, and students are encouraged to contact Bentley IT for guidance regarding supported AI technologies, appropriate use, and potential security concerns.
Last updated: 09/09/2026