Katie Harbath

@katieharbath.bsky.social

Work at intersection of tech and democracy. Fmr Facebook. Newsletter writer: anchorchange.substack.com

I wrote Disrupting Politics for many reasons, but one was that I didn't want anyone else telling my story. This book is dedicated to the people in the arena with me: trust and safety, campaigns, tech policy, doing hard work that rarely gets the nuance it deserves. open.substack.com/pub/anchorch...

Why I Wrote Disrupting Politics and Why It's Not Just About Facebook

An insider's look at the decisions nobody had a playbook for and why I had to tell it myself

open.substack.com

Bluesky's new CEO Toni Schneider on the platform's reputation for being a liberal bubble: "Yes, we definitely want that to change. It is already changing. It certainly wasn’t designed to attract one specific group of people."

In Meta’s Q1 earnings call, they focused on how AI is making our feeds more personal. It’s making me think about how we manage the tradeoffs between what's optimized for each of us individually might not be automatically good for us collectively. anchorchange.substack.com/p/ai-optimiz...

AI Optimized for Me. Optimized for You. But What About Us?

What Meta's Q1 earnings reveal about the personalization tradeoff — and what we might lose along the way.

anchorchange.substack.com

Thrilled to publish my latest article BAN COOKIE BANNERS: A CASE STUDY IN TECH REGULATION with the Harv. J. of Law & Tech! A cri de cœur to end this worse-than-useless regulatory compliance regime and let something that actually protects users take its place. papers.ssrn.com/sol3/papers.... 1/

Ban Cookie Banners: A Case Study in Tech Regulation

Few experiences on the modern internet are as universally reviled as the cookie banner. They clutter websites with pop-ups, interrupt user flow, make informatio

papers.ssrn.com

There's a major gap in content safety tooling: classifiers typically only score complete text. When you're working with generative AI, "complete text" means the user already saw it. That's too late. So we built a streaming classifier that we're releasing today! Here's what we did and why. 🧵...

I partnered with the Rainey Center to poll 1,008 Americans on AI, media consumption, and tech policy heading into 2026. The toplines looked coherent. The crosstabs told a completely different story. Here are the six contradictions that stood out: 🧵1/3

This week, I'm publishing my 2026 AI & Election Tech Forecast—focused intelligence on how major platforms are navigating the collision of AI, elections, and content moderation. Up first: Meta. What I'm watching: 💰 $65 million in corporate spending on pro-AI candidates 1/3

Politics is heavy. The pace of AI is exhausting. We're at the start of what might be the most consequential four-year cycle for tech and politics we've seen yet. In my latest newsletter, I break down what's actually different this time. open.substack.com/pub/anchorch...

How to Read Tech Platforms During a Never-Ending Election Cycle

The 2026 midterms start a four-year period where platforms, politicians, and AI reshape each other. Here's what to watch when no one knows what's coming next...

open.substack.com

America's digital empire has a trust problem. U.S. hyperscalers' future may depend less on the commercial deals they strike and more on their commitment to protect the democracy that made their global dominance possible. My latest for @wpr.bsky.social www.worldpoliticsreview.com/us-hyperscal...

America’s Digital Empire Has a Trust Problem

One year into Donald Trump’s presidency, foreign nations are increasingly skittish about their dependence on U.S. tech companies.

worldpoliticsreview.com

"Which events should I attend in 2026?" Here's my framework of what I do before, during and after attending something. Later this week, I’m sending subscribers a full list of the calendar of events I’m tracking in 2026. Sign up to be the first to get it! anchorchange.substack.com/p/how-i-deci...

How I Decide Which Events Actually Matter

A look inside my process for navigating conferences, summits, and what comes after

anchorchange.substack.com

Over the weekend I used Zentropi to build a labeler that blocks requests to use AI to undress or sexualize real people. The labeler itself took maybe 30 minutes to a first solid draft, with another 30 minutes of tweaking. In my testing it's got an F1 of .98 on real examples pulled from X. 🧵 1/6