[solatic]

> After all, in an open ecosystem like Matrix, it's possible someone will fire up a buggy/malicious client and inadvertently compromise a room.

Isn't this a case for building in some kind of system whereby clients can be signed and have their signatures revoked by their creators or, for lack of a better word, ostracized by the wider community? Sort of like a web of trust model, but for clients, not just users, to make it more clear when somebody is joining with an untrusted client and perhaps allow moderator control over whether to allow untrusted clients to join.

[Arathorn]

Possibly, but this is heading into seriously DRM territory. one would need to be running the app in some kind of secure enclave that could attest to the authenticity of the app (e.g. via SGX on Intel). There's something a bit unsavoury about saying that "only truly official signed apps are allowed to participate in this open network", and it gives a huge amount of power to those responsible for the secure enclave/trusted computing stuff. (There's also the approach that djb mentions in https://twitter.com/hashbreaker/status/732912508089032706)

It's possible that just relying on social mechanisms may be enough to discourage people from running known evil apps (similar to educating users not to install malware today, or do trusted operations with cybercafe computers, or whatever). Effectively, the verification process when going and explicitly trusting a new device needs to explicitly prompt the user to consider where on earth this app came from, and if it should be trusted.

The only alternative is really DRM, which just feels wrong.

[solatic]

>There's something a bit unsavoury about saying that "only truly official signed apps are allowed to participate in this open network", and it gives a huge amount of power to those responsible for the secure enclave/trusted computing stuff.

Maybe it's a bit naive, but isn't that what federation is supposed to solve? People who are more security-paranoid can forbid clients which don't have the highest security certification, and operators who aren't so diligent will be fine with signed clients being run on untrusted hardware.

I mean... is there any open-source software being developed today which enforces key security in secure hardware enclaves? Verifying the GPG signatures on binary packages is "good-enough" for most operators. Build reproduceability will help to further reduce trust of unverified hardware.

It seems to me the job of the protocol, and baseline/recommended UI/UX, is merely to help users make informed decisions. Security is a spectrum, and if signed clients improve security (while not fraudulently representing itself as perfect or near-perfect security, if it were running on trusted hardware), then that's a net benefit to the open network.

[Arathorn]

I may be missing something, but how do you prove that an app is running a trusted codebase? I know of no PGP clients for instance that sign messages to try to prove that they were sent from a trusted app (as opposed to a trusted user). The only way I know of to do this would be to hook into a trusted execution environment of some flavour like Intel's SGX or Apple's Secure Enclave, to let effectively the chip vendor sign off that you are running an official app installed by official means. You /could/ do this, but you are putting a lot of trust in the secure enclave implementation and those controlling it, and essentially putting all your eggs in one basket. It might also lure users into a false sense of security: just because a user is using an appstore signed copy of an app doesn't mean that app is actually trustworthy or bug free. And it would also artificially discriminate against legitimate apps which aren't part of a trusted execution environment, which seems dangerous - and effectively promoting DRM at the expense of FOSS.

This certainly needs more thought :)