CroissanthologyIt's better to personally do your part in neutralizing bad actors than to fuss about your cybersecurity stack

It's better to personally do your part in neutralizing bad actors than to fuss about your cybersecurity stack

also on Substack

Personal cybersecurity is offense dominant (for now)

In principle, I'm partial to the idea of privacy. GrapheneOS sounds cool, and so does a home server for storing one’s 1.8TB of archiving content. Being in full control of your digital wake solves one of the oddest arrangements we have with large companies: “store my data and make it readily available, and in exchange I'll give you effective control of it”??1 The US government seems starkly unaligned to the American people at the moment, and the less information they have on me, the better. The cost of sifting through mountains of information in the name of a given goal has plummeted, so anyone with access to my data could do many more useful things with it.2 “Truesight” by LLMs could have non-linear gains, in the sense that “recognizing this text as Kelsey Piper's” could be the first step in a capabilities buildup that ends up with the ability to do more with your data than “merely” singling you out among billions after sampling a single paragraph (for example, persuading you).

At the same time, security reasons for privacy are slightly dumb. The NSA has your encrypted data if it’s ever gone through tapped optic fiber wires, and they're storing it in a vault to decrypt later. Superpersuasion by AIs e.g. creating a Xhosa prophecy would probably work on everyone given how similar human minds are, so why bother specifically protecting your own data. Offense is the true way to ensure bad people can't do bad things to you: vote for the right government, work on AI alignment, advocate for policies, support EU-style laws requiring copies of your data be transmitted upon request (even if your account is banned), etc. There are still good reasons, including principle, to build a complicated home server contraption that avoids Google; and doing so is getting cheaper time- and skill-wise. It may even be possible to do it perfectly (see link below). But for security, you're probably better off thinking about the far harder problem of ensuring nothing is coming to get you in the first place.

LLM cybersecurity risks in the long run are probably exaggerated18 Feb 2026Read →

As the cybersecurity saying goes, your threat is either Mossad or not-Mossad: don't even bother trying to defend against the Mossad. (Unfortunately, acquiring a Mossad is getting cheaper—but the saying still holds!)3