# The Existential Risk of Knowing Too Much (Especially About People)

By Wes Pomeroy. Published April 10, 2026. https://wespomeroy.com/writing/ai-and-personal-data/

Recently I've been getting into some deep LLM stuff. Real deep. If you don't know about Claude (Desktop/Cowork/CLI) - it's a game changer. Enables use cases far beyond the chatbot-as-fancy-search-engine paradigm.

This technology is so powerful, so exciting, that my life has essentially become one long Wikipedia rabbit hole. So many ideas, so many projects, so many use cases - there's not enough time in the day. At work I lean in hard to my classic day job, proactively and reactively. But in every spare moment I'm working to use new capabilities and radically transform the way we work. That sounds grandiose and self-important, but it's not. The wave is starting to curl and I'm desperately trying to surf to shore before it crashes on top of me. That's the white-collar paradigm shift we're all staring down.

With that - there are now endless topics to write about. I'm going to start doing that here, and sharing the journey.

**My topic today - the existential risk of knowing too much (especially about people).**

One of the main things I've been working on, at home and at work, is doing deep analysis on my entire email corpus. Twenty-plus years of personal Gmail at home. Seven-plus years of work Gmail at work. Everything is in there. People. Projects. Successes. Failures. Best practices and wisdom. Unfulfilled dreams and roadmap items. Interpersonal dynamics. Good approaches and bad approaches. All value that's ever been added. All opportunities. It's like discovering you've been keeping a detailed journal for twenty years without realizing it - and suddenly having a way to read the whole thing at once.

Never before has it been possible to retroactively analyze such a rich source of knowledge and wisdom. But with that comes a risk I don't think we've really had to face as a species: knowing too much.

*I think I can bring that point home in one thought exercise: imagine if you and your partner could read each other's minds.*

Scary, right? There's a reason we filter what we expose to the outside world. We "pick our battles." We "choose our words." We selectively forget. We remember and forget things in ways that serve us emotionally more than they accurately reflect the historical record. Our personal corpus may include all of our virtues and wisdom and wins. But it also includes all of our sins, our foolishness, and our failures.

I made the mistake of asking my personal system to look back on 20 years of emails and give me a top-ten list of "Areas of Opportunity" in terms of being a better person. I took one look at the list and filed it away. It was too much. It was too true. Too much at odds with my self-image as a generally good, empathetic, and righteous person. I'm sure most of it was accurate. I'm just not able to look in the mirror with that clarity, aperture, and magnification.

My first attempt at work, however, was wildly successful. The insights my system delivered have been profound. Life changing. Has already improved relationships with people I've worked with for over a decade.

But in working on this, my spider-sense is exploding with concern. This is like nuclear fission. It can power the planet or destroy it. We generally keep negative and unconstructive stuff out of the written record, but it's there - especially in chat and IM. What about sensitive information? Compensation details? Performance issues? Things said, regretted, and best forgotten? There could be some really unconstructive stuff that probably should stay buried in the historical record.

My conclusion is this: parsing and leveraging rich people data requires very clear guardrails.

1. When in doubt, go broad and general, not specific, when analyzing granular people data.

1. Build hard, structural, contract-based filters to keep sensitive people data out of your pipeline.

1. Be explicit and transparent about your ethical and privacy boundaries, and codify them into your AI system.

1. Train your system to be conservative in its assumptions and comfortable saying "I don't know" or "not enough evidence" - and generous in the assumptions it does make about people.

1. Don't think you can "stop" this kind of analysis or put the genie back in the bottle. You cannot un-invent nuclear technology. You can't stop music piracy. Risk mitigation needs to be grounded in what is actually controllable.

1. Work directly with HR, Ethics, and other experts in this space. AI catalyzes expertise rather than replacing it.

*This is the path I plan to walk.*