Re: Who writes the ACLs on your pod?

Hi,

 

The sentence in bold below is what I would consider also lands in the ‘agents’ space (Alice gives access to Bob the human, the builder or the bot) … the answer I’ve got is that, at the moment, this is an unchartered territory (I already asked OIDC, link below, another post coming mid next week).

 

https://mayakaczorowski.com/blogs/ai-agent-authentication

 

And it would probably look like a combination of authentications + claims + rights management. A journey somehow like this:

 

- Alice “delegates” to BtB, (there needs to be a way to “narrow” the available actions to BtB - that’s part of the “delegates” that needs to be agreed upon)

- BtB tries to access a resource <BtB, action, resource> 

- PEP checks that Alice has delegated to BtB (there is a “relationship” between Alice and BtB) to allow for the actions

 

(Delegate used in the broadest sense for any action).

 

For the rest of your scenarios, not sure if I follow, but it could be helpful to think that you have several actors:
The owner (“Alice”)
Pro-active Agents (like any of the Bob’s)
Passive Agents (applications triggered by Alice)
The data
 

At least in ODRL you can have agents (called “parties”) in multiple roles (and they can be extensible to more). In your model Alice is the owner (“assigner”) and any agent will take a different role (non-assigner) over the asset (“the data”).

 

Your vocabulary of actions then gets decomposed to CRUD ‘orders’ (for example, “modify” over “calendar” allows for “writing”/”updating”/”deleting” the entries, “modify” over “agenda” means you can delete the calendar altogether, etc).

 

Regards,

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: Michiel de Jong <michiel@unhosted.org>
Date: Monday 3 February 2025 at 15:37
To: Joshua Cornejo <josh@marketdata.md>
Cc: <public-solid@w3.org>
Subject: Re: Who writes the ACLs on your pod?
Resent-From: <public-solid@w3.org>
Resent-Date: Mon, 03 Feb 2025 15:36:42 +0000

 

To put it another way, I'm looking at the scenario where there is just one user, but there are multiple apps. For instance, the user is switching from one app to another, the 'plausible exit' scenario.

 

We should of course also try to implement the scenario where one user ('Alice') shares something with another user ('Bob') but for that use case it's even harder than just the question of the title of this thread.

 

Additional questions that arise there would be who should decide which app Bob should use for this (Alice? Bob? Alice's pod provider?) and we also don't have a good way for Bob to get notified that something was shared with him. The obvious way would be to send a Linked Data Notification to Bob's inbox, but I'm not aware of any Solid server implementation that actually pings its users when something hits their inbox. And of course also the additional question - if Alice gives Bob access to a resource on her pod, how can she express the purpose and the use restrictions in a way that she can trust Bob will adhere to. So the question I'm asking in this thread is far from the only question ahead of us. :)

 

Cheers,

Michiel

 

On Mon, 3 Feb 2025 at 15:46, Michiel de Jong <michiel@unhosted.org> wrote:

Hi Joshua,

 

Yes, that is indeed important to get clear. I'm not talking about GDPR or ODRL or DPV here, although they can obviously help. I'm purely talking about the Solid protocol and the Client-Client specs of the Solid project here - in particular type indexes as a way to know what type of data is in each document and/or container on the pod, as a basis for access decisions.

 

> authorisation by third parties

 

Who are the first, second and third parties in this case?

 

The way I'm modeling it here, there are four parties involved:

* the pod (storage+idp) owner/user

* the pod (and with that, its provider)

* the app launcher (and with that, its provider)

* the app (and with that, its provider)

 

The app launcher edits the ACLs in response to instructions from the pod owner.

In SAI, this is the authorization agent editing the registries the achieve the same.

 

The pod owner makes the decision

The app launcher translates the decision into edits in ACL documents

The pod (provider) hosts the ACL documents alongside the data in the pod, and also enforces the policy described by them

The app gets access to the data, or not. :)

 

HTH!

Michiel

 

On Mon, 3 Feb 2025 at 15:29, Joshua Cornejo <josh@marketdata.md> wrote:

Michiel,

 

Just to make things clearer and create the right context, OIDC definition of “Authorize” is not the same as the one Harsh/me are talking about.

 

OIDC uses authorisation for coarse grain elements of the identity (claims connected to an account, things like “email” or “uuid”) during the process of authentication.

 

 

(Harsh correct me if wrong) DPV refers to the language used to describe authorisation by third parties to access *ANY* data (parts of authentication or parts of the data inside app) according to specific controls (processes, legal or otherwise).

 

ODRL is a language to describe rights management – from CRUD to sophisticated permissions/prohibitions/obligations (and any extended taxonomy of rules).

 

Regards,

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: Michiel de Jong <michiel@unhosted.org>
Date: Monday 3 February 2025 at 14:19
To: Joshua Cornejo <josh@marketdata.md>
Cc: <public-solid@w3.org>
Subject: Re: Who writes the ACLs on your pod?
Resent-From: <public-solid@w3.org>
Resent-Date: Mon, 03 Feb 2025 14:18:26 +0000

 

Thanks for your replies! That gives a lot of context and vocabulary for this discussion.

 

Nevertheless, I think the open questions that remain are:

1) For our ACP+OIDC-based Solid servers, will we restrict the root ACL to one or more specific app store(s) or app launcher(s) like I did here?

2) For our WAC+OIDC-based Solid servers, what should we do? Have a really stern warning like I did here? Not ideal of course! We want to do better. But what is the way forward?

3) We have the SAI spec which may actually one day replace  ACLs altogether, or become an answer to the question of how we edit the ACLs on our pod, but none of our Solid servers have adopted it yet. Can we somehow work together to change that, so that we can try it out on at least one Solid server?

 

If we apply one of these three options then we can begin to define a minimal version of interoperability between Solid apps. I have spent considerable effort on all three options over the past years, and have recently ended up picking the first one.

 

I did some more demo building and my "app launcher" now has three bookmarking apps in its "app store": https://github.com/pdsinterop/launcher-exploration/blob/957a0dacd5baa8e4d9bef1681c363195767bd3e8/app-permissions.js#L1-L14

 

It now demonstrates actual interop between real Solid apps! :) https://www.youtube.com/watch?v=I48DW0F4bX4&t=269s 

 

So, achievement unlocked, in that sense. :) But even with that, more questions remain:

 

A) Should we try to form a consensus and gain traction on one of the three options? Or should we keep the options open to allow for more diversity and experimentation?

B) Should access be edited along the lines of the private and public type index, or can we do better? More precisely, do we want to stick to the type indexes approach and try to build an ecosystem of interoperable apps with that, or do we want to develop something better than type index based, or do we want to do both of those things at the same time?

C) Assuming we want there to be multiple independent implementations of Solid app launchers/stores apart from the rudimentary one I put together for this demo, and we don't want to enforce one centralised app registry on them, how can we nonetheless help them manage their app registries in a collaborative way, and can https://github.com/solid/catalog help there?

D) Is scoped access for Solid applications actually a goal in itself we want to achieve, or should we not work on scoped access unless we also work on other dimensions of fine-grained consent (such as purpose, agreement to policies, audit, etc).

 

Cheers,

Michiel de Jong

 

On Sat, 1 Feb 2025 at 10:33, Joshua Cornejo <josh@marketdata.md> wrote:

I agree with Harsh, an ISO standard follows an architecture along these lines:

 

 
Everyone is connected, networks will get faster and lower latency
Follows the principles of triples (actor, action, asset) and coexists with emerging standards
With enough decorations, you can also implement temporary caching (e.g. “access for one-day” type) to reduce network/latency/trips
The audit is not necessarily the responsibility of the owner of the container and might not even be the responsibility of the PDP/registry provider
External PDP allow for changes and enforcement at scale - one place that affects large amounts of components
External policy registries allow for consistency and reduction of cost of implementation, training, maintenance
Lowers barrier of entry (a PEP component is built once – used anywhere) to create more complex behaviours using public policies
 

Regards,

 

___________________________________

Joshua Cornejo

marketdata

smart authorisation management for the AI-era

 

From: Harshvardhan Pandit <me@harshp.com>
Date: Saturday 1 February 2025 at 00:38
To: <public-solid@w3.org>
Subject: Re: Who writes the ACLs on your pod?
Resent-From: <public-solid@w3.org>
Resent-Date: Sat, 01 Feb 2025 00:38:01 +0000

 

Hi.

A 'registry' (is this different from an 'app store' as in the Apple on or a 'package repository' as in the Linux ones?) is definitely the correct step forward IMHO for practical governance and management of apps. Two relevant papers that are helpful for governance of apps for things like policies, notice/consent/contracts, and also things to consider regarding lock-ins etc.

 

1) Making Sense of Solid for Data Governance and GDPR https://doi.org/10.3390/info14020114 (see Section 5.2 in particular)

2) Using Patterns to Manage Governance of Solid Apps https://ceur-ws.org/Vol-3636/paper5.pdf (See fig.1 on pg.5 to summarise)

 

For "access scope" - it would be better to not reinvent terms where possible. If we're talking about a description of data used/processed by the app, then it goes beyond 'access' (see article 1 above). Currently, this information is contained within a document called a 'privacy notice' where it also includes information such as who is doing what, what controls are available to start/stop something, and how to exercise rights and requests. But we definitely shouldn't repeat that rubbish modus operandi of drowning people in large legalese documents. So instead, we can think something like the Privacy Labels in the Apple App store which, but much more detailed and useful - in machine-readable form. This way no one has to 'read' anything and 'agents' have the information to support decision making in context. This follows from the existing philosophy of what Solid aims towards IMO.

 

In terms of having a 'language' to do all of these - DPV https://w3id.org/dpv is fully capable of expressing all pertinent details, and can be extended to add granular/arbitrary things like specific data categories, 'purposes', technical measures, and actors/roles for Solid. ODRL is a standard to express agreements/permissions/prohibitions/etc. - so it provides the framework for stating something is a request, a user policy, an agreed use of app/data, etc on top of DPV. So the only thing remaining is creating a spec and getting folks to use/improve it.

 

Regards, Harsh

 

On Wed, 29 Jan 2025, at 09:30, Michiel de Jong wrote:

Hi all,

 

I wanted to share this 9-minute screencast about an experiment we originally did in 2019 and recently updated to the latest spec version: https://www.youtube.com/watch?v=5Q1nUmGdaXE

 

In this example, the launcher app "knows" that Poddit is an app that requires access to instances of type 'bookmark', through a hard-coded registry.

 

Within the Solid project, and also maybe cross-fertilising with other related projects on this, we could do a lot of improvement on how such an app registry would work in the real world, and on developing a language to express access scopes such as "all your things of type Bookmark".

 

Cheers,

Michiel