Lars Gyllenhaal
Writing
3 min read

Why I Don't Use Two-Factor Authentication on My Personal Systems

This requires some preliminary clarification, because the title will be read uncharitably by people who think security advice should be universal. It should not be. Security advice that applies to everyone is advice calibrated for the median threat, which means it is wrong for most people and catastrophically wrong for others. I am not the median threat.

The standard case for two-factor authentication rests on a specific attack model: an attacker has obtained your password — through breach, phishing, or reuse — and is attempting to authenticate as you from a location or device that is not yours. In this scenario, a second factor breaks the chain. The attack fails. This is correct reasoning, and for most users with most accounts, it applies.

My personal systems are not exposed to this attack. They are not publicly accessible. Authentication occurs over physically controlled channels or, in a small number of cases, over authenticated tunnels where the second factor is the tunnel itself. An attacker who has my password still cannot use it, because the authentication surface is not reachable. Placing a TOTP code requirement in front of an interface that cannot be accessed from outside a physically secured network adds friction to my workflow without adding any meaningful security property. It is a control designed for a threat that does not apply.

The error people make when encountering this kind of reasoning is to conclude that I am discounting the value of 2FA generally. I am not. I am discounting its value in a specific, bounded context where the threat model does not support it. This is the same reasoning that leads me to not wear a bulletproof vest in a supermarket. Not because armor is ineffective. Because the threat is not there.

What this points to is a broader problem with how security advice is transmitted. A control that is appropriate in one context becomes dogma, gets applied universally, and then the entire discourse about it collapses into "use it" versus "don't use it" — losing the only part of the conversation that matters, which is: use it when your threat model supports it, because you have one, and it is specific.

Most people do not have a specific threat model. For them, 2FA is probably correct. Most people also do not disable 2FA for considered reasons — they disable it because the TOTP code is annoying. I am not making that argument. I am making a different argument, and the fact that the same conclusion can be reached via laziness does not make the reasoned version wrong.

I use two-factor authentication where the threat model supports it. On my personal systems, it does not. This is not a revelation. It is an application of first principles to a specific context. I mention it because first-principles thinking is apparently rare enough to be worth documenting.