I Uploaded My Family's Tax Return to ChatGPT. Then I Read the Privacy Policy.
I needed a quick answer about my RRSP contribution room
It was tax season. I had my T4, my RRSP receipt, and a CRA notice I didn’t fully understand. I wanted to know whether I’d overcontributed. So I did what millions of people do: I uploaded the documents to ChatGPT and asked.
ChatGPT gave me a clear, helpful answer in about four seconds. It pulled the relevant numbers from my T4, cross-referenced the RRSP receipt, and explained the CRA notice in plain language. I was impressed.
Then I read the privacy policy.
What I actually sent to OpenAI
My T4 slip alone contained:
- My full legal name
- My Social Insurance Number
- My employer’s name and payroll account number
- My total employment income
- My CPP and EI contributions
- My union dues and pension adjustments
The RRSP receipt added my financial institution, account number, and contribution amount. The CRA notice included my address, my assessed income from previous years, and my RRSP deduction limit.
Combined, these three documents give anyone everything they need to file a fraudulent tax return in my name, open credit in my name, or redirect my CRA benefits to a different bank account.
And I had just sent all of it to a server in San Francisco.
What OpenAI’s privacy policy actually says
I went to OpenAI’s privacy policy and terms of service. Here is what I found, quoted or paraphrased directly:
On data collection
OpenAI collects “content and metadata” of inputs and outputs, including files uploaded to the service. This includes the full text content of uploaded documents.
On training
By default, your conversations and uploaded files can be used to train OpenAI’s models. You can opt out in Settings > Data Controls, but this setting is not enabled by default, and most users never find it.
On retention
Even with training disabled, OpenAI retains your data for up to 30 days for “safety and abuse monitoring.” The policy does not specify exactly what monitoring occurs or who has access during this period.
On jurisdiction
OpenAI is a US company. Your data is subject to US law, including the CLOUD Act, which allows US government agencies to compel access to data held by US companies, regardless of where the data subject lives. Your SIN, your income, your employer: all accessible under a US court order.
What Google Gemini says
Gemini’s terms are similar in practice. By default, conversations are used for “product improvement.” Google’s privacy policy is broad enough to cover most uses of your data. The opt-out exists but requires navigating through multiple settings pages.
The core problem is the same: you are sending your most sensitive financial documents to a company whose business model depends on processing as much data as possible.
Why tax documents are different from meeting notes
When someone uploads meeting notes to ChatGPT, the worst case is that a competitor learns about a project timeline. Annoying, but recoverable.
When someone uploads a tax return, the worst case is identity theft that takes years to unwind. The CRA’s identity theft process requires filing police reports, contacting credit bureaus, and waiting months for resolution. During that time, someone else may be collecting your GST credits, your child benefits, or your tax refund.
Leaked: project timeline | Leaked: SIN, income, employer, bank details
Impact: competitive disadvantage | Impact: identity theft, fraudulent filings
Recovery: days | Recovery: months to years
Frequency of change: weekly | Frequency of change: annually (SIN never)
What I do now instead
I still need AI help with tax documents. The questions haven’t changed: Did I overcontribute to my RRSP? What does this CRA notice mean? Can I claim this home office expense?
But now I upload those documents to Archevi instead of ChatGPT. Here is what happens differently:
I upload my T4 and RRSP receipt to Archevi. The documents are stored on an encrypted server in Toronto, Canada.
Before any AI processes my question, Archevi automatically detects and replaces my SIN, my name, my employer, and my account numbers with anonymous surrogates. The AI never sees the real data.
I ask my question: “Did I overcontribute to my RRSP this year?” The AI answers using the document content, but works with anonymised values. My actual SIN never leaves Canada.
The answer includes citations pointing to the exact lines in my documents. I can verify the AI’s reasoning against the source material.
Archevi hard-blocks eight categories of sensitive data before any AI processing: Social Insurance Numbers, credit card numbers, bank account numbers, passport numbers, driver’s licence numbers, health card numbers, date of birth, and phone numbers. These are never sent to any external service, regardless of AI provider settings.
The real question is not whether ChatGPT is useful
It is. The summarisation, the plain-language explanations, the ability to cross-reference documents: all genuinely valuable. The question is whether the convenience is worth sending your complete financial identity to a US company that retains it for at least 30 days, may use it for training by default, and operates under a legal framework that allows government access without your knowledge.
For meeting notes, the answer is probably yes. For your family’s tax return, I would argue it is not.
Upload a tax document to Archevi and ask it a question. Watch how the anonymisation works in real time. Your SIN stays in Canada. Free plan, no credit card required.
