[OpenID-Specs-eKYC-IDA] Usage of scopes and purposes when requesting verified claims

Joseph Heenan joseph at authlete.com
Mon May 25 10:10:48 UTC 2020


Hi all,

I’m a little late into this conversation, but I’d like to share some recent experience.

The use of the ’scope’ shorthand in OpenID Connect Core is not quite as tightly defined as you might like - the spec:

https://openid.net/specs/openid-connect-core-1_0.html#ScopeClaims <https://openid.net/specs/openid-connect-core-1_0.html#ScopeClaims>

says:

> The Claims requested by the profile, email, address, and phone scope values are returned from the UserInfo Endpoint, as described in Section 5.3.2, when a response_type value is used that results in an Access Token being issued. However, when no Access Token is issued (which is the case for the response_type value id_token), the resulting Claims are returned in the ID Token. <https://openid.net/specs/openid-connect-core-1_0.html#UserInfoResponse>
 <https://openid.net/specs/openid-connect-core-1_0.html#UserInfoResponse>
This doesn’t go as far as saying “the email MUST NOT be returned in the front channel id_token for response_type=code id_token” (there’s also language elsewhere that most people miss suggesting the email claim maybe shouldn’t be returned in the front channel if it would be available via a back channel method).

For example, I recently discovered that several OPs, including at least one very large one, always return ‘email’ in all id_tokens (as well as the intended location in the userinfo response) when requested by scope=email. This is not disallowed by the Connect spec, but returning the more sensitive information involved in eKYCs responses in front channel id_tokens or in back channel id_tokens (that may be shared with third parties etc) would probably be unwise and/or not unexpected by the RP - could potentially lead to to PII being held in unexpected places or unintentionally shared.

I’d perhaps suggest adding a privacy or security consideration along the lines that as a minimum any agreement to use scope shorthands (particularly when they’d apply across a whole ecosystem and hence RPs may not closely check exactly how each OP behaves) should make it clear which locations the resulting claims must/must not be returned in.

Cheers,

Joseph



> On 8 May 2020, at 08:22, Torsten Lodderstedt via Openid-specs-ekyc-ida <openid-specs-ekyc-ida at lists.openid.net> wrote:
> 
> Hi Achim,
> 
> sounds reasonable. I will create PRs.
> 
> thanks,
> Torsten. 
> 
>> On 5. May 2020, at 11:08, Achim Schlosser <achim.schlosser at enid.eu> wrote:
>> 
>> Hi all,
>> 
>> 
>> Moving this out of in-line comments:
>> 
>> Scope Usage: 
>> 
>> In general there is nothing in the spec that would prohibit RP and OP to agree on sets of claims/trust frameworks that they would expose via simple scope requests. I would suggest to add some general language around this and leave it there, scopes are by definition only shortcuts to claim requests.
>> 
>> Proposal:
>> 
>> 5.x Requesting Verified Claims using Scope Values
>> 
>> As defined in (REF 5.4 Core Spec https://openid.net/specs/openid-connect-core-1_0.html#ScopeClaims) scopes can be used to request that specific sets of information be made available as Claim Values. This specification defines no additional predefined scopes. OPs MAY define scope values to request verified claims. 
>> 
>> Purpose:
>> 
>> As with similar topics I would strongly encourage to not venture into Data Protection details in any form with the specification, but rather soften the language and not go beyond what is already written in the core spec. 
>> 
>> Core spec defines this - https://openid.net/specs/openid-connect-core-1_0.html#Consent - Just leave it with that. The usage of the word consent in the core spec is already problematic as it is now with GDPR in mind, so it can only get worse.
>> 
>> Proposal:
>> 
>> Remove any MUST language in relation to the purpose and also MAY language in case it relates to unilateral action of the OP.
>> 
>> 
>> Best
>> 
>> Achim 
>> 
>> On 01.05.20, 18:26, "Openid-specs-ekyc-ida on behalf of Torsten Lodderstedt via Openid-specs-ekyc-ida" <openid-specs-ekyc-ida-bounces at lists.openid.net on behalf of openid-specs-ekyc-ida at lists.openid.net> wrote:
>> 
>>   Hi Jaap,
>> 
>>> On 1. May 2020, at 15:23, Jaap Francke <jaap.francke at iwelcome.com> wrote:
>>> 
>>> Hi Torsten,
>>> 
>>> Thanks for your reply. My comments are also inline.
>>> 
>>>>> 
>>>>> https://openid.net/specs/openid-connect-4-identity-assurance-1_0.html
>>>>> 
>>>>> The "OIDC for identity Assurance” specs introduce a number of claims, in addition to the ones specificied by OIDC core.
>>>>> In OIDC Core, there are 2 mechanisms to request claims:
>>>>> - one is by usage of OAuth-scopes, (section 5.4 of Core spec)
>>>>> - one by usage of the “claims” parameter in the request. (section 5.5 of Core spec)
>>>>> Section 5 of the OIDC—identity-assurance specs indicate the usage of the claims pararameter. This may suggest it’s the only mechanism to be used in the context of Identity Assurance.
>>>>> My view is that also the ’scope’ mechanism should be supported and may even be preferred for certain use cases.
>>>>> Please enhance the specification to be more explicit about usage of scopes as a means to request verified claims.
>>>>> 
>>>>> Usage of ’scope’ to request verified claims makes sense to me because in a typical “identity landscape” the requested claims do not vary on a request-by-request basis, but instead are a reflection of an RP’s functionality which can be rather static. By pre-defining/configuring scopes at the OP (potentially in terms of ’essential’, ’trust_framework’, etc..), the scope is essentially a profile of the claim request. This would not only simplify the protocol implementation both at the RP and OP side, but it would also make it easier for the OP to make the authorisation decision whether or not the RP will be granted the requested verified claims. 'It is at the discretion of the OP to decide whether the requested verification data is provided to the RP.’ Making this decision is much easier when a request includes a scope as a reference to a predefined profile of requested end-user claims and associated verification data. My suggestion is to consider including such a mechanism in the specifications.
>>>> 
>>>> What do you have in mind? I agree a certain RP or deployment could define scope values as short cut to request certain verified claims. However, the number and value range of verification elements + the number of end user claims creates is so big I cannot envision how the spec could define concrete scope values.
>>> 
>>> Yes, so what i had in mind is that the spec would acknowledge the practice of defining a ‘custom' scope value as shortcut to request certain claims. The current spec only mentions the ‘claims’ request parameter as mechanism to request claims, which I see as somewhat ambigous The spec could mention both mechanisms as valid options and leave it to the the implementor to make a choice which mechanism is more appropriate. 
>>> My preference is the ’scope’ approach, since it fits into what i see as a typical situation where the interaction between RP and OP has a more ’static’ nature. 
>>> I could imagine the spec even stating perceived pros&cons of both approaches. Claims approach gives a lot of flexibility to vary on a request-by-request basis. The scope approach is easier in situations where there is a ’static’ interaction between OP and RP; i.e. an RP may always want to receive the same set of claims, for the same data processing and purpose, etc.
>> 
>>   I assume the interaction can be static for a certain RP with a certain OP. This would call for RP-specific scope values. Have you ever seen this in practice? I haven’t. 
>> 
>> 
>>> 
>>>> Additionally, the spec no longer defines values for trust frameworks, verification methods and evidence, i.e. there are unknown values from a spec perspective.
>>>>> 
>>>>> My second observation is about the optional ‘purpose’ parameter in a claims request. The specs state that the OP MUST display this purpose in the respective user consent screen(s) in order to inform the user about the designated use of the data to be transferred or the authorization to be approved. If the parameter purpose is not present in the request, the OP MAY display a value that was pre-configured for the respective RP.’ 
>>>>> This part of the specification assumes or seems to imply that the user gets a consent screen every time that verified data is requested. I think that this is not realistic:
>>>>> - it is common practice that certain consents are included in a Privacy Policy (PP) and/or Terms of Service (ToS). The user gives his consent once and the consent is persisted at the OP for future purposes.
>>>> 
>>>> Are you referring to ToS/PP of the RP or the OP?
>>> 
>>> Good question. I think it could be either one or maybe even both. My understanding of the European GDPR is that the OP may need the user’s consent to provide data to the RP and the RP needs a legal ground to collect a person’s data. This is assuming that RP and OP are different legal entities / Data Controllers. Different scenario’s can also be possible where OP and/or RP (from a legal perspecrive) could be Data Controller or Data Processor. Moreover, i think legal entities can even delegate the responsibility to get a user’s consent (if consent is the legal ground that is applicable)
>>> 
>>> 
>>>>> - if not in PP or ToS, a consent screen may be presented only ‘once’ and the user’s consent is taken to be valid for the next 6 months (as an example)
>>>>> - in some situations there are other legal grounds/reasons/purposes to request the claims and verification data. Besides ‘consent’, the European GDPR indicates other legal grounds (contract, vital interest, legal obligation, ..). When an RP asks for a verified claim, it may do so based on a legal obligation and the user’s consent would not be needed.
>>>>> - Privacy regulations aim at situations where personal data is exchanged between legal entities (data controllers, data processors). From an IT perspective, the OP and the RP may be operated by the same entity (for example an Insurance company as data controller that is both the data controller for the OP and various RPs) so consent would not apply at the moment of requesting the claim. The consent should already have been given at the moment the personal data was provided to the OP.
>>>>> My proposal would be that the ‘purpose’ could be a string (as per current specification) or a reference to a ‘purpose’ that has already been established somewhere. This approach makes it easier for the OP to make the decision to provide the requested data or not.
>>>> 
>>>> I agree the text is too strong. The purpose should be used to further inform the user in case a user consent screen is required. I will try to change the text accordingly.
>>> 
>>> Extending my previous replies, my suggestion would be that the eKYC standard would not ‘enforce’ or assume whatever particular legal set-up of responsibilities, but instead facilitate any legal context and/or interpretation.
>>> I think the way to do so would be to include the possibility to include one or more references to types of data processing (‘purposes’) that should somehow be legally covered.
>>> How this ‘legal cover’ is arranged for may vary per implementation:
>>> - Is the OP Data Controller or Data Processer? Is the RP Data Controller or Data Processer?
>>> - Does the intended data processing require the user’s consent or does one of the other legal grounds apply?
>>> - Did RP and OP agree on some delegation about getting the user’s consent (if needed)?
>>> In this way the protocol can facilitate any set-up in a transparant way, without making assumptions on legal reponsibilities at either OP and/or RP.
>> 
>>   What does this mean to protocol constructs? Can you please suggest concrete changes/text?
>> 
>>> 
>>> 
>>>>> 
>>>>> Furthermore, it could be considered to extend the list of verification methods with the ones defined by NISTIR 8112.
>>>> 
>>>> You mean „Document Verification”, “Record Verification”, “Document Verification with Record Verification”, “Proof of Possession”, “Probabilistic Verification”, „Not Verified”? We could add this to our Wiki. May I ask you to add the values?
>>> 
>>> Yes that’s what I meant. I need to familiarize with the wiki.
>> 
>>   You may find the existing definitions at https://bitbucket.org/openid/ekyc-ida/wiki/identifiers.
>> 
>>   Let me know if you want to contribute changes, the I would give you the respective permissions. 
>> 
>>   best regards,
>>   Torsten. 
>> 
>>>> 
>>>> -- 
>>>> Openid-specs-ekyc-ida mailing list
>>>> Openid-specs-ekyc-ida at lists.openid.net
>>>> http://lists.openid.net/mailman/listinfo/openid-specs-ekyc-ida
>>> 
>> 
>>   -- 
>>   Openid-specs-ekyc-ida mailing list
>>   Openid-specs-ekyc-ida at lists.openid.net
>>   http://lists.openid.net/mailman/listinfo/openid-specs-ekyc-ida
>> 
> 
> -- 
> Openid-specs-ekyc-ida mailing list
> Openid-specs-ekyc-ida at lists.openid.net
> http://lists.openid.net/mailman/listinfo/openid-specs-ekyc-ida

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs-ekyc-ida/attachments/20200525/27ba31dd/attachment-0001.html>


More information about the Openid-specs-ekyc-ida mailing list