You couldn't pay my enough, I need social connection

Next time this happens, folks, @blackskyweb.xyz often works as an alternative client and AppView while Bluesky is down

Really feels like the open Internet is falling apart, with bsky as a rare bright spot.

The more I think on it, the more I realize how much the confused sheriff problem already haunts us as a society. Every scapegoat is a victim of a confused sheriff.

All this recent talk of matriarchy makes me a bit wary. I reject that women are intrinsically good at care and men intrinsically bad. That biological essentialism is terribly counterproductive, and no one knew that better than black feminists.

White women have the most skill issues for historical reasons.

Black men worked the fields.
Black women did the most care work.
White men managed the external economic affairs.
White women...looked pretty and had babies.

A truth white feminism doesn't acknowledge enough is that the legacy of care it ascribes to white women was historically more the labor of black women, and it shows because black women fairly consistently have the better more intersectional feminism.

THREAD: I've now analyzed @massgovernor.bsky.social's proposed social media legislation and ...

... it is somehow even worse than the House bill that passed last week (which was already one of the worst bills on this issue we've seen in the US.)

You might wanna sit down for this. Here goes: 🧵

What would protect you even more, is a PDS with a private audit log letting you prove that you never once accepted the delegation, but that's a future innovation :)

The confused sheriff could be literal, and we really want to make sure that sheriff is not confused.

If you never accept the malicious delegation with a take record, you could conceivable use that as evidence to prove your innocence. You could prove that you never accepted delegation into the incriminating context, and therefore had no involvement.

IANAL, but this could even be of legal consequence. Someone could try to implicate you in illegal activity by delegating authority to you with a give record. Capability trees protects you from this by requiring a take record from you to complete the social contract and activate the capability.

Where the confused deputy grants authority poorly at execution time, the confused sheriff attributes accountability poorly at audit time. Both are confused because the authority isn't clear enough, and both are solved with more legible authority.

This threat model emerges specifically when designing a social capabilities system like capability trees. It's a social threat model that enters the picture because the system is social. It's new because prior capabilities systems have not been social in this manner.

While designing capability trees, I theorize a novel threat model: The confused sheriff problem.

I think I've been letting AI do my thinking a little too much. Claude floated the social confused deputy problem, but that was never quite it. It's really about the confused sheriff problem, which has more to do with audits than execution. To be fair though, I did have to invent that vocabulary.

The sheriff is on the outside working out who is accountable based on who has authority. The confused sheriff misattributes accountability because the authority isn't legible enough. This roughly mirrors the confused deputy, granting authority poorly also because the authority isn't legible enough.

It emerges specifically in social capability systems: public, persistent, identity-bound delegation where the ledger is readable as evidence. Prior capability systems were not social this way, and therefore didn't have this attack surface. Bilateral consent closes it before it opens.

I came up with a cute name for this new attack surface: The confused sheriff problem.

In terms of social threat model, there could be reputational damage from someone delegating you into something you don't want to be involved with. Protocol level mutual consent means your unassociation is visible to all, as the absence of a take record from you.

Took a while to give this more thought. The function this bilateral structure serves is more social than technical. In the ATProto context, it requires both sides to consent to the relationship. Google Docs has a similar feeling feature, with invite emails feeling like a delegate's opt in step.

What would proper citation of oral traditions look like?

Thanks for clarifying. Good design!

Short expiry on a signed invocation goes a long way to protect you for sure. Adding voluntary scope attenuation goes even farther. Layered security for the win!

Right. And once you sign it, you've produced a signed UCAN, right? If it was still somewhere in the system after you're done, someone could use it, because it's already signed.

How does UCAN use the private key to do that, if not by signing?

I'm picturing a situation where the attacker obtains a lingering signed UCAN still on the machine somewhere after you leave, but still doesn't have your private key. In which case, they can impersonate you with the signed UCAN, but cannot request more permissions.

Tbh, the voluntary scope attenuation threat model is less contrived. Suppose you've granted me write access to something, but I'm using a library computer that the next user could try to hack, so I only want to accept read access until I get home. It's about reducing my own attack surface.

I'd wager those are a good foundation for technical skills

On a technical level, it allows for voluntary scope attenuation - taking a smaller scope than what you're given.