Of the 12K tweets where users prompted Grok ~3k were sexual in nature. Of those requests, Grok complied 91% of the time.

Surprisingly, image editing in general had a slight reduction in compliance (83%).

When is your data yours?

For enterprise, training on customer data is prohibited by default across all providers analyzed.

For consumers, the standard is an opt-out model, where data usage is assumed unless settings are manually changed. Uniquely, xAI trains on data even if you are logged out.

Next, indemnification determines the financial burden of a third-party lawsuit.

Enterprise agreements generally offer "Mutual Indemnification," sharing the burden of third-party lawsuits.

For consumers, this is nonexistent - meaning the user actually pays to defend the company.

For context, mandatory arbitration clauses waive a user's right to public court trials and class action lawsuits.

It means issues are mediated by a third-party - not a court.

In these data, OpenAI alone mandates binding arbitration for all users, regardless of tier.

If anyone is interested in these data, I am happy to provide upon request.

Given the Dem dominance, I wanted to look deeper at Republican co-sponsors. Surprisingly, 69% of co-sponsors on Rep AI bills were Dems.

That is, when Reps introduced bills, they tended to do so solo. But Dems had a stronger tendency to build coalitions within and across party lines.

As you can see, Adam Schiff dominated with 3x the centrality of 2nd place Amy Klobuchar. Brian Fitzpatrick was the only Republican even in the top 12.

This means Democrats, especially, have a strong amount of within-group co-sponsorship.

Notably, only 18% of connections cross party lines.

To find who bridges the partisan divide, I built a co-sponsorship network where legislators are connected if they co-sponsor the same bill.

To do this, I used a metric called betweenness centrality which measures how often someone sits on the path between others. A high score means they're a bridge

The first plot shows that of the 140 bills introduced, 1 became law.

You can see a post-ChatGPT surge with an obvious partisan split. Dems created bills more focused on regulation and privacy, whereas Reps pushed trade and workforce.

Next I was curious to see which members of Congress drove them.

128 AI bills with full text were pulled (2019-2024) and categorized by policy area using TF-IDF (Term-Frequency Inverse Document Frequency). This method finds text with similar word usage to make distinct groups.

Then I pulled co-sponsorship data to build a network and see who is connected to whom.