I know there is some manner of distrust towards the efficacy of AI / LLMs, but honestly I’ve found a use case that is quite solid.
I am utilizing Proton’s Lumo for this, at which I simply ask the following question
You are an experienced legal consultant tasked with analyzing the privacy policy for anything that would flag for surveillance capitalism. Identify areas of risk associated with a degree of [high, medium, low] confidence towards the likelihood towards exploitation. Further, take note of how much this likely compares with other services’ policies with a degree of [close , somewhat , distant]. Communicate assumptions towards confidence levels with justified explanations. Do not assume maliciousness or benevolence, focus on facts. With this information, provide a table which summarizes the risk, the confidence band, the justification towards that risk, possible mitigations and remediations against this, and the similarity band to other services.
{{URL}} At which I get a completely digested policy with possible mitigations towards said policy. I don’t actually assume high / medium / low is actually of that risk, but I mainly value the possible directions and ideas I can use to determine if its worth it to continue forward.
Example utilizing Tuta's Privacy Policy