Toledo Labs has reviewed the legal privacy terms for major LLM providers such as OpenAI and Anthropic. To use Anthropic as an example I downloaded their commerical terms into a pdf and uploaded it to Toledo1 to ask this document questions rather than read the entire thing manually to get to the bottom of what I wanted to know as quickly as possible. First, I warmed up the model by asking it to summarize the commerical terms.
Then I asked about the privacy terms on API inputs and outputs and got the following response from Toledo1:
- Data Ownership:
- The document states that “Anthropic agrees that Customer owns all Outputs, and disclaims any rights it receives to the Customer Content under these Terms.”
- Customer Content is defined as including both Prompts (inputs) and Outputs.
- Data Usage:
- Anthropic explicitly states that it “may not train models on Customer Content from paid Services.”
- Data Protection:
- If personal data is submitted to the Services, Anthropic’s Data Processing Addendum applies.
- Anthropic commits to processing data only in accordance with applicable data protection laws and on the documented instructions of the Customer.
- Data Deletion:
- Upon termination or expiry of the Agreement, Anthropic will, within 30 days: a) If requested, return a copy of all Covered Data or provide self-service functionality for the Customer to do so. b) Delete all other copies of Covered Data processed by Anthropic or its Sub-processors.
- Security:
- Anthropic commits to implementing and maintaining reasonable and appropriate technical and organizational data protection and security measures.
- Sub-processors:
- Anthropic may use sub-processors to process data, but must enter into agreements with them that are no less protective than Anthropic’s obligations under the DPA.
- International Transfers:
- The document includes provisions for international data transfers, including adherence to Standard Contractual Clauses where applicable.
Item 2 is the ahaha moment lets review this:
Anthropic explicitly states that it “may not train models on Customer Content from paid Services.”
Then I uploaded all of Anthropics legal documents into Toledo1 and asked:
Last question, is it fair to say after reading all of their terms the only certainty for privacy would be through paid services?
In conclusion, there is no such thing as free inference. As a business owner I would never use it and risk my IP. If you don’t pay in dollars you pay with ideas.