AI at Bigelow: Frequently Asked Questions
Updated
These answers respond to the questions and concerns staff raised in the October 2026 AI survey. For the full rules, see the AI acceptable use policy at support.bigelow.org/ai-policy.
Privacy and data
Will AI companies train their models on our data?
Not with Bigelow's accounts. Anthropic does not use conversations from Bigelow's Claude organization account to train its models, and Bigelow Chat (chat.bigelow.org) only uses models configured for zero data retention. The risk is real for free personal accounts that haven't opted out of "model improvement" — if you use one, follow the opt-out steps at support.bigelow.org/ai-policy.
Where are my conversations stored?
- Claude (Bigelow account): on Anthropic's servers, protected by our agreement with them. This is the same kind of trust we already place in Google, NetSuite, 1Password and Blackbaud.
- Bigelow Chat (chat.bigelow.org): on Bigelow's own server. Each request is sent to the model through Amazon Bedrock and is not retained after it is answered.
Can IT or my supervisor read my chats?
Supervisors have no access to your chats. IT can see account and usage information, such as who has an account and how much it is used. We do not read anyone's conversations, but we want to be clear about what is technically possible:
- Claude: as the account owner, the head of IT can export and search the organization's conversations. We would only do this when legally required or as part of a security or compliance investigation, with approval from the executive team.
- Bigelow Chat: conversations are stored on Bigelow's own server, so IT staff with server access could technically read them. We don't, as a matter of principle.
As with Bigelow email and shared drives, treat work AI accounts as work tools rather than a private personal space.
What data can I not put into AI tools?
Current policy at support.bigelow.org/ai-policy
Intellectual property and funder rules
If I put data/thoughts into AI chat does that make it public?
No. Entering data or ideas into Bigelow's Claude account or Bigelow Chat does not publish them. Your conversations are not shared with other users, not used to train AI models, and not visible to anyone outside Bigelow. It is no different from saving a file to Google Drive or sending a work email: the information stays private to Bigelow under our agreements with those vendors.
Two things can change that, so avoid them for anything sensitive: using a free personal account that hasn't opted out of model training, and sharing a chat link with people outside Bigelow.
Who owns what the AI produces?
Under Anthropic's commercial terms, Bigelow keeps ownership of what we put in and what comes out. You remain responsible for checking that output before using it, and for crediting sources it draws on.
My funder or project has rules about AI. Which applies?
The stricter rule always wins. Some funders limit AI use; for example, NIH prohibits reviewers from using AI tools on applications they are reviewing. Check your award terms and data agreements, and ask the VP of Research if you're unsure.
Accuracy and using AI well
AI makes things up. How can I trust it?
Treat it like a fast, capable colleague whose work you always check. It can state wrong facts, invent citations and make subtle errors in code or analysis. To reduce the risk:
- Give it the source material (a paper, dataset, document, or trusted website) and ask it to work from that.
- Ask it to cite its sources, then open and check them.
- Test code and spot-check calculations before relying on them.
You are responsible for anything you use, the same as with any other tool or reference.
Checking its work takes longer than doing it myself. Is it worth it?
Sometimes not, and that's fine. AI tends to save the most time on first drafts, summarizing long documents, writing and debugging code, and reformatting. It saves the least on tasks that need precise, verified facts. The upcoming trainings will share examples of where colleagues have found real time savings.
Environment and ethics
What about AI's energy and water use?
It is a fair concern, especially for Bigelow. Data centers that run AI use significant electricity and water, though the footprint of a single everyday chat is small compared with training the models. You can reduce your share by:
- using AI where it saves real effort, not by habit
- choosing a smaller, faster model for simple tasks when you have the option
- avoiding repeated regeneration of the same answer
Can we trust these companies?
No vendor is risk-free, which is why we put agreements in place and decide which data goes where. Our approach relies on contracts and settings rather than goodwill alone. Bigelow Chat gives an option that keeps conversation history on our own systems.