[OpenID] TransparencyCamp and OpenID (U)

Fri Mar 13 16:49:48 UTC 2009

I don't see why the delegation feature of OpenID would particularly interfere. One could be doing exactly the same delegation with SAML.

B exploiting the "heavy" nameid modules in SAML, one could do it rather better than openid does it, not that anyone does - to my knowledge  (At Rapattoni, we accomplish what openid delegation achieves by non SAML means, seeing as our SAML server doesn't do nameid. It uses "legacy" associative-mappings that realty has had deployed for 10 years.)

Most of the rules you commented on were about the signals and the security handshake , not the quality of user authentication. I see no real difference between SAML.authcontextclass and openid.pape.

The NIST criteria doesn't impose any trust models - they do not exclude self-asserting parties, for example (the openid UCI model).

If trust issues are not out of scope, they should be. Assurance criteria should not be advocating or biasing any administrative elements of operational policy .

...Otherwise, this  process will do to SAML/openid ...just what it did to PKI.

From: general-bounces at openid.net [mailto:general-bounces at openid.net] On Behalf Of Breno de Medeiros
Sent: Friday, March 13, 2009 9:28 AM
To: Paul Madsen
Cc: OpenID List
Subject: Re: [OpenID] TransparencyCamp and OpenID (U)

On Fri, Mar 13, 2009 at 3:42 AM, Paul Madsen <paulmadsen at rogers.com<mailto:paulmadsen at rogers.com>> wrote:
Full NIST LOA requirements for 'assertions' (NIST uses the term in the inclusive sense and not specific to SAML) are laid in Section 10.3.2 of http://csrc.nist.gov/publications/drafts/800-63-rev1/SP800-63-Rev1_Dec2008.pdf

Focusing on Level 2, the combined assertion requirements of LOA 1 & 2 are

1a) All assertions recognized within this guideline must indicate the assurance level of the initial authentication of the Claimant to the Verifier using the former's long-term token(s). The assurance level indication within the assertion may be implicit (e.g., through the identity of the Verifier implicitly indicating the resulting assurance level) or explicit (e.g., through an explicit field within the assertion.)

Implicit indication could be achieved by public documentation.

1b)  it must be impractical for an Attacker to manufacture an assertion or assertion reference that can be used to impersonate the Subscriber. If the direct model is used, the assertion which is used must be signed by the Verifier or integrity protected using a secret key shared by the Verifier and Relying Party, and if the indirect model is used, the assertion reference which is used must have a minimum of 64 bits of entropy.

Shared secret key => implies that the OP should not accept negotiations over HTTP. Even DH does not bind the shared key to the verifier because there is no authenticity.
Response_nonce should have at least 64 bits of entropy.

1c) Assertions shall be specific to a single transaction, and, if assertion references are used, they shall be freshly generated whenever a new assertion is created by the Verifier. In other words, assertions and assertion references are generated for one time use.

Response_nonce, if properly implemented, would guarantee this. Both the RP and the OP (in check_authentication mode) should verify the nonce for uniqueness.

1d) all assertions sent from the Verifier to the Relying Party must either be signed by the Verifier, or transmitted from an authenticated Verifier via a protected channel. In either case, a strong mechanism must be in place which allows the Relying Party to establish a binding between the assertion reference and its corresponding assertion, based on integrity protected (or signed) communications with the authenticated Verifier.

Implies that OP check_authentication should be over SSL only.

1e) assertions that are consumed by a Relying Party which is not part of the same Internet domain as the Verifier must expire within 5 minutes of their creation. Assertions used within a single Internet domain, including assertions contained in or referenced by cookies, however, may last as long as 12 hours.

That means that the RP should reject response_nonces older than five minutes. Implies some level of clock skew adjustment that OpenID does not provide.

2a) assertions must specify whether the name of the Subscriber is a verified name or a pseudonym.

Assuming it to be a pseudonym unless otherwise specified by extensions would satisfy this.

2b) assertions shall be protected against manufacture/modification, capture, redirect and reuse.,

Reuse is prevented by the signature inclusion of the return_to URL, op_endpoint. Signature also prevents against modification. Protecting against capture could be understood as a mandate for POST, since GET assertions can be compromised through the browser history prior to their expiration.

2c) assertions, assertion references and any session cookies used by the Verifier or Relying party for authentication purposes, shall be transmitted to the Subscriber through a protected channel which is linked to the primary authentication process in such a way that session hijacking attacks are resisted

This is not applicable, as OpenID does not rely on common-domain cookies.

2d) To protect assertions against manufacture, modification, and disclosure, assertions which are sent from the Verifier to the Relying Party, whether directly or through the Subscriber's device, shall either be sent via a mutually authenticated channel between the Verifier and Relying Party, or equivalently shall be signed by the Verifier and encrypted for the Relying Party.

Again, implies that check_authentication should be performed over SSL only.

2e) All assertion protocols used at Level 2 and above require the use of Approved cryptographic techniques. As such, the use of Kerberos keys derived from user generated passwords is not permitted at Level 2 or above.

SSL would likely meet such standards.

In the surface, it would seem that OpenID could achieve such standard with minor tweaks of the protocol. However, this perception is misleading. The standards was developed for non-user-centric systems where the IDP has out-of-band mechanisms to bind identifiers to real users. In a user-centric system where users can delegate arbitrary identifiers to the OP, one would expect it to be necessary to develop an additional set of criteria by which delegation can be assured to a minimum level of integrity protection. Today delegation can be achieved reasonably easily via phishing attacks that may appear relatively benign (do not ask user to enter password credentials, for instance).

paul

Brett McDowell wrote:
To (what I think is) Chris' point about changing/up-leveling the discussion out of a "OpenID vs. SAML" or "OpenID is insecure" sidetrack, and to (what I think is) Johannes' point about capturing the security requirements that we can base a new work stream on... I humbly suggest we look at the requirements our friends in the US Government (and others) are requiring now and are confident they will continue to require regardless of (and independent of) private sector innovation.  I'm specifically referring to their requirement to compliance to the NIST defined levels of assurance (LOA).

If I were to peer into my crystal ball I would probably see that most of the compelling applications that governments will open to 3rd-party credentialed citizens are likely to be set to LOA 2 or LOA 3.  How about we shift this conversation to focus on how OP's can offer OpenID-based services to their users that achieve government recognized compliance to LOA 2 (and soon after, LOA 3)?

Some work on this has already been *started* but not progressed in earnest.  Project Concordia has some use cases that look at this problem space.  The Identity Assurance Framework looks at how any particular credential service can achieve LOA 1 through LOA 4.  What we don't have is any analysis of what an OP could achieve with OpenID 2.0.  Knowing this will provide a clear gap analysis of what we have vs. what we need. We can base our deliberations on these hard facts.  I can only believe this will be more productive than... actually I don't see any alternative to this approach if we are serious about making progress.

Next Steps?

Since I work closely with the primary US Government agency in charge of procurement (which is quite important to the issue of what/which IT infrastructure Federal agencies deploy), and because they already require our certification for all federation technology prior to being considered by agencies for procurement (this is all SAML and/or PKI  today -- just to clear up any confusion about the usage of SAML in eCitizen applications), and since they are working closely with us to adopt our credential assessment framework for LOA 1-2 in the very near future (and LOA 3-4 down the road), I would be happy to talk with them about co-hosting a kick-off event to drill into this issue as it relates to OpenID specifically.   I assume they will be interested.  They, like I, would like to see citizens be able to use whatever private sector credentials they "already have" to access government applications.  If those are OpenID's, then lets make sure those OpenID's are going to be acceptable to these federal Relying Parties (who knows, we might learn something that helps us win more RP adoption in other markets as well).

Thoughts?

Brett McDowell | +1.413.652.1248 | http://info.brettmcdowell.com

On Mar 12, 2009, at 9:05 PM, Peter Williams wrote:

I've been playing with dynamic SAML metadata modes recently. Instead of reading a signed XRDS file, peers dynamically  sign SAML metadata files. A SAML metadata file is only XML, and has extensions:  one of which could include an XRD or 2 (and thus get signed XRD off the ground). You can look at the SAML endpoints as funky XRD services, all expressed in their own markup.

The best thing that ever happened to SAML2 was openid - as [most of] the SAML crowd have largely got off their ultra high horse and been forced to match openid in simplicity and effectiveness (or be made irrelevant).

From: general-bounces at openid.net<mailto:general-bounces at openid.net> [mailto:general-bounces at openid.net] On Behalf Of Chris Messina
Sent: Thursday, March 12, 2009 3:59 PM
To: Johannes Ernst
Cc: OpenID List
Subject: Re: [OpenID] TransparencyCamp and OpenID (U)

On Thu, Mar 12, 2009 at 4:37 PM, Johannes Ernst <jernst+openid.net<http://openid.net>@netmesh.us<http://netmesh.us>> wrote:

On Mar 12, 2009, at 9:46, Ben Laurie wrote:
I agree that all of SAML is way too large a pill to swallow but
there's no reason subsets that are usable cannot be defined, and,
indeed, have been.

I would love it if somebody was actually starting a working group (in Apache they would call it "incubate") that would propose all the gory details of a "more secure" form of OpenID that still fits into the decentralized, discovery-based OpenID architecture. Only then can we tell what may or may not be the better approach.

+1.

I'm not the person to do it, simply because I don't have the background, but I'm really no longer interested in the discussion that "OpenID isn't good enough for high-value transactions". Tell that to the Japanese who are already pushing payments over/with OpenID.

If it's got security issues, as Johannes said, we should collect a list of them as issues and go about finding suitable solutions, best practices or explanations for WONTFIX-type resolutions.

The reality is, people will use OpenID in all kinds of cases that we can't be able to anticipate; SAML (again, from what I hear - being mostly SAML ignorant) is that SAML requires sysadmins to maintain and more and more organizations want to move away from that kind of model.

HTTP succeeds because (among many, many other reasons) you can jack up your service to things like S3 once you realize that you're not really saving any money as a small to mid-sized organization running a server farm. The same thing applies to user authentication and security over time, where it'd be nice to have a fairly straight-forward, interoperable protocol for doing authentication. From what I've heard, I don't question the sophistication or technological pedigree of SAML; it's just that when I hear people say that OpenID isn't secure enough, it's like saying we should all switch to RDF because it's OMG-so-much-better.

I don't doubt that either are, if your primary concerns in life are centered around the problems that these technologies solve; but for folks for whom security isn't necessarily their only responsibility, and they no longer have an IT department or staff to manage a SAML solution and look to OpenID as a potential answer - saying that OpenID isn't secure enough doesn't seem to be an answer to the question "well, then what do I use?"

How do we get from where we are today to a point where OpenID really does solve a large number of security problems - if only because it hopefully means fewer passwords in use overall - and fewer unique accounts to maintain? Look, let's take another look at this: it isn't just "is OpenID by itself secure?" It's: "given how OpenID changes the makeup of and behavior in the overall ecosystem, are we now more secure than we were before?" In general, I think the answer is yes, though of course good security necessitates vigilance and constant improvement.

To Johannes point: are we capturing these needed improvements anywhere, and if not, can we begin to?

Chris

--
Chris Messina
Citizen-Participant &
 Open Web Advocate-at-Large

factoryjoe.com<http://factoryjoe.com> # diso-project.org<http://diso-project.org>
citizenagency.com<http://citizenagency.com> # vidoop.com<http://vidoop.com>
This email is:   [ ] bloggable    [X] ask first   [ ] private
_______________________________________________
general mailing list
general at openid.net<mailto:general at openid.net>
http://openid.net/mailman/listinfo/general

________________________________

_______________________________________________

general mailing list

general at openid.net<mailto:general at openid.net>

http://openid.net/mailman/listinfo/general

________________________________

No virus found in this incoming message.

Checked by AVG.

Version: 7.5.557 / Virus Database: 270.11.10/1996 - Release Date: 11/03/2009 8:42 PM

--
Paul Madsen
e:paulmadsen @ ntt-at.com<http://ntt-at.com>
p:613-482-0432
m:613-282-8647
web:connectid.blogspot.com<http://connectid.blogspot.com>
[cid:image001.gif at 01C9A3C0.2D1096C0]<http://feeds.feedburner.com/%7Er/blogspot/gMwy/%7E6/1>

_______________________________________________
general mailing list
general at openid.net<mailto:general at openid.net>
http://openid.net/mailman/listinfo/general

--
--Breno

+1 (650) 214-1007 desk
+1 (408) 212-0135 (Grand Central)
MTV-41-3 : 383-A
PST (GMT-8) / PDT(GMT-7)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-general/attachments/20090313/8c746257/attachment-0002.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.gif
Type: image/gif
Size: 22342 bytes
Desc: image001.gif
URL: <http://lists.openid.net/pipermail/openid-general/attachments/20090313/8c746257/attachment-0002.gif>