How we use AI
Last updated: October 31, 2026
This page is not legal advice.
What the AI does
Sidesheet uses AI to read the pages of the tax returns you upload and pull out the line items for your financials. It does not give tax, valuation, or accounting advice (Terms §3.4).
Which model, and where
- Provider: Google Cloud, through Vertex AI, using Google's Gemini models and Anthropic's Claude Haiku, which Google serves on Vertex AI under the same Google Cloud terms.
- Location: the United States. The model runs in Google's US multi-region (
us), which keeps processing inside the US; your stored files stay inus-central1. Our code accepts only US locations and refuses Google's "global" setting, which could route requests outside the US. - We use Google's paid enterprise service, under the Google Cloud terms, not a free consumer tool.
What we send to the model
The content of the pages of the returns you upload: the text and numbers on them, including any names or taxpayer ID numbers that appear there. We do not remove numbers the financials need, because that would make the financials wrong. We send nothing else about you or your firm.
No training on your data
- Us: we do not use your data to train or fine-tune any AI model (Terms §2.3).
- Google: Google's terms say: "Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction." (Google Cloud Service Specific Terms, §18 "Training Restriction"). We have not given that permission and will not.
What Google keeps, and for how long
From AI processing: nothing. We have turned off Google's request and response logging, Gemini's input caching is off, and Google exempted our project from keeping prompts for abuse monitoring (granted 2026-09-28; confirmed for Gemini and Claude Haiku on 2026-10-04). Your files go to the model inside each request and are not stored by the AI service.
Your files themselves are stored in Google Cloud Storage, in our own project in the US, encrypted with our own key. Google holds them only as our storage provider (Subprocessors). A file is deleted when your firm removes it, or 7 years after upload. After that, we can still recover it for 7 days, and then it is gone for good.
You check every number
AI can make mistakes. So:
- Every extracted number links to the page of the return it came from.
- Values the system is unsure about are flagged for your review.
- You, the professional, are responsible for checking the financials before relying on them (Terms §3.4).
Fixing reading errors
When the service detects a likely error in reading an upload (for example, a failed or incomplete extraction, a check that doesn't tie, or a value your user corrected), or you report one, our staff may look at that upload and its results, only as needed to fix the error (Terms §4.2). A fix may change how the service reads similar returns for all customers, but we keep no copy of your data to make it. Apart from that, we don't review or reuse your returns to improve extraction. If we ever offer a broader program, it will be off unless you turn it on, and we will announce it first (Terms §2.4).
A note for firms on AI disclosure laws
Some states regulate how professionals use AI with clients. For example, Utah's SB 226 (2025) requires a licensed professional to disclose AI use when a person receiving their services interacts with conversational generative AI in a "high-risk" interaction. In our service, only your firm's staff use the software. The business being valued never interacts with our AI, and our AI is not a chat tool. We think that makes such rules unlikely to apply, but your obligations depend on your state and your engagement. Please check your own.
Questions
privacy@sidesheet.co · security@sidesheet.co · See also our Security page and Subprocessors.