15 Jun 2025 Search

Update: The reply from Verizon was:

You have successfully submitted a request to download information. We were not able to complete your privacy request due to a valid exemption under the CPRA or we couldn’t find any records associated with the information you provided.

I know they have me in their system, so this looks like it’s probably an error on the agent side.

Also I’m going to make sure that agentic AI can reliably do RtKs for me before I start trusting it with other form-filling tasks like booking travel.

Status: The agent got part way through, but no success.

Original post:

Pretty soon after I posted an agentic AI benchmark, already built, I got Mark Hinkle’s newsletter, How to Actually Start Using AI Agents Today. Perfect timing—thank you, universe! Now I have no excuse not to test the most promising AI agents and see if they will help with some of the complex processes that you have to go through to do a right to know (RtK) under CCPA or some other state privacy law.

So I made an account on Manus and tried one of my go-to RtKs, Verizon. I like RtKing them because they have a bunch of LOL-worthy marketing insights (inferences, which under the law must be disclosed) about my family, or at least they did last time. Verizon is a good first RtK for people getting started with this exciting(ish) hobby.

This company is not too hard to RTK, but they do have some forms to get through, so hard enough to be a realistic test for AI. Here’s the transcript with my contact info redacted. tl;dr: it worked (mostly).

My first prompt was a little disappointing. I provided what I thought was the right form URL, and Manus got stuck. But I tried again. This time I supplied the privacy policy URL and told Manus to look in the California section, and it got through.

Only one error. I provided a phone number but it filled in n/a until corrected. But as soon as I supplied the right number it did the right thing.

The problem with RtKs compared to other software tests is that one person can only RtK the same company every so often. So I’d be really interested in seeing what happens when other people do variations on this one. Don’t feel bad about overloading Verizon with RtKs—if they can afford to pay off the former Twitter to show ads to bots then they can afford to do these.

  1. Make a Manus account.

  2. Prompt it with your contact info and the URL of the privacy policy to start with (don’t go down the dead end that I did) https://www.verizon.com/about/privacy/full-privacy-policy

  3. See if it can get to the success screen, and if the RtK data shows up.

Screenshot of the end of an AI agent session

I know a lot of people are still doubtful that AI is making much progress in doing original work, but solving RtK mazes isn’t really that. RtK mazes are a class of problems that somebody already designed, like solving CAPTCHAs.

More: personal AI in the rugpull economy The biggest win from personal AI will, strangely enough, be in de-personalizing your personal information environment. By doing the privacy labour for you, the “agentic” AI will limit your “addressability” and reduce personalization risks.