Dario Amodei of Anthropic wrote an essay titled “We Must Pace the Frontier” in the last couple of days. He essentially argues that AI could facilitate harms that we do not intend, especially in warfare, biological weapons, and cybersecurity attacks. The New York Times has a related article: “Top A.I. Leaders Call for Slowing Down A.I. Development“: In an essay, the head of Anthropic laid out the rapidly advancing capabilities of artificial intelligence and said there needed to be greater safety controls across the industry.
I had to think back to the post below and why I wrote it a year ago. I never published it, but I think some of the ideas still hold.
Exfiltration of data, where large quantities of data are pulled from systems and used in unintended ways, is a very real risk with the systems we build today. A very senior engineering leader I worked with previously in my career left one of the frontier labs to work on this very question. Knowing all of the risks, we must really be thoughtful about how much data we centralize. It is one of the reasons I have worked on the concept of a Personal Neutral Archive that can be owned by a person and not necessarily connected to these larger botnets that Dario warns about.
I wrote the post below on 10/18/2025.
[TLDR: There are ways to work with AI safely and some ways are safer than others. Consumer health data should stay in consumer’s hands. We can enable care and research at scale without having to compromise human privacy. The technology already exists to use AI on device with encryption in distributed architectures.]
It is clear from my writing that I am generally an Artificial Intelligence optimist, and that I want to see it safely deployed to the world.
With that goal in mind, I want to help a body of work unfold around the technology that exists today to keep consumer health protected when interacting with AI models. We can use AI at the edge. We can use smaller AI models on edge devices. We can use larger models in data centers in privacy preserving ways.
We can achieve a view of the patient’s lifetime of data only for the person and their clinicians to see. We can enable research at scale while preserving that privacy. We do not need massive data repositories to achieve the social benefits we seek.
Literature has always offered us a window into the human condition. In the Lord of the Rings trilogy, Tolkien paints a picture of a technology (the ring) so powerful its allure is consuming. We see this today with AI.
In The Fellowship of the Ring, Gandalf firmly refuses to take the One Ring from Frodo. He says, “No! Do not tempt me! I would wish to turn it to good, and it cannot be used so!” He is aware that his own desires to help and protect the world could get twisted and lead to domination. He believes that the Ring would corrupt him into believing his actions are for the greater good, when in reality they create danger.
The overall moral of the story is that power should not be concentrated in the hands of any one entity. It shows us, too, that it’s possible to be mindful of risks while proceeding forward in fellowship.
Would an AGI be too powerful? Well, certainly if we feed it data on every living biological organism we’re not doing ourselves any favors. Better to keep the good stuff a little more highly distributed, if you ask me.
There are plenty of ways we can both make rapid scientific advancements and manage risks. In fact all adoption of scientific achievement involves managing risks well. I don’t get on an airplane thinking that nobody checked it out, but it can go super fast.
When you get medical care today, we already use millions of clinical technologies at the edge to help understand what’s happening. We manage risk on those devices 24 hours a day. We have interoperability standards to move data where it needs to go. You can get a copy of your data. Can we make that process more frictionless? Sure. Can we learn novel things in new ways? Definitely. Can we start all over and engineer AI-native systems? Yes.
I hope that those who have worked with all kinds of technologies for many decades can help society see that there could be multiple efficient methods to help consumers and that some methods, while expedient, create greater risk for society. We have brilliant teams creating new solutions every day, and we also have solutions for reducing the risks.
If you ask me, your power (and your data) should be hidden in your pocket like Frodo understood.
——————————————
I wrote this post this morning before reading the BBC article below entitled, “Is this how AI might eliminate humanity?” I reread my post after reading the article, and I think my thoughts hold true. Let me know your thoughts 🙂
https://www.bbc.com/reel/video/p0lwdg8y/is-this-how-ai-might-eliminate-humanity-