One of key challenges in federated authentication network is the establishment of trust between an identity provider (IDP or OP) and relying party websites (RP). In the real world, contractual agreements provide a simple out-of-band mechanism to effectively bind two parties into a trust relationship.
When it comes to federated identity networks, peer to peer contracts between many identity providers and a myriad of relying party websites do not provide for a scalable process. Therefore, open federated networks need a trust assurance framework to bootstrap trust between the three parties (the user, the OP and the RP).
The basic idea is that if an OP can be certified to comply with a set of industry best practices, the RP should be able to enter into open identity exchange where both the websites and the consumers are reasonably protected. Of course, a pragmatic trust assurance framework should be flexible enough to support different levels of assurance based on the transaction risk and value.
For low assurance Web federation where large brands such as email providers and major social networks dominate as OPs, certification may seem overkill, unless of course, the federation is built on open principles stating that any OP meeting the standard should be able to participate. For high assurance identity, such as payment networks, financial networks or eHealth record exchanges, certification is primordial. In fact, in such environments, both the OP(s) and the RPs need to be certified.
The NIST guideline for electronic authentication is often referenced in the community as a good model for any identity trust framework. The NIST guideline defines four levels of insurance for e-authentication. Each level is deemed appropriate depending on transactional risks. Tiered levels of identity assurance are essential to any pragmatic trust framework. Set the bar too high and deployment becomes impractical. Set the bar too low, and the bad guys will have a ball. Justifiably, the NIST guideline provides a solid starting point.
Nevertheless, one needs to observe that the framework may be too narrowly focused on user credentialing and credentials strength to provide a complete answer. Open Identity systems cannot ignore the reality of today’s Web vulnerabilities, threats and exploits that feed identity theft around the globes such as man in the browser exploits, session hijacking or Web vulnerability driven exploits like mass SQL injections. A trust standard also needs to go beyond security and address the major consumer concerns and political challenges of privacy. When it comes to trusting identities, security, privacy and anonymity are intricately intertwined. Trust in a federated identity Web mandates a holistic approach that looks not only at user authentication but also takes into account the current state of desktop exploits, Web site compromises and most importantly establishes clear and enforceable privacy protection guidelines.
Trusting the OP/RP Websites: web security & business authentication
For low and medium assurance identity transactions, it seems to be that both the OP and RP website security would need to be asserted. There I think, one can learn from Internet security standard such as PCI. Even though the standard is far from being perfect (a euphemism, perhaps), it provides a shared base of security requirements for all websites to engage into ecommerce and securely handle credit card information. If one believes that consumers will require for their personal identity the same level of security as for their credit card, the parallel can be useful. The OP website should then be scanned for network security vulnerabilities; Ports should be closed.
Network services should not run outdated or un-patched software; the OP should not be vulnerable to common Web exploits such SQL injections, cross-site scripting (XSS), or Cross-Site Forgery requests (CSRF). For web application vulnerabilities, the OWASP standard that identifies the top 10 Web vulnerabilities provides a useful reference. In addition to security assessment, a set of security best practices should be required. For example, the OpenID profile retained by the federal pilot already specifies that SSL should be part of the deployment profile. Verifying the authenticity and legitimacy of the organization behind the OP is as important as verifying the security of its website. There, a proven model that the industry could re-use is the EV business authentication standard. EV certification already defines a strong process for vetting organizations and it is already widely used across the industry.
Trusting the user: beyond identity verification and credentials
As mentioned, NIST will provide the foundation for user trust assurance (both for runtime and initial authentication of end users). Equally important, however, is to consider that Internet threats have significantly evolved since the NIST framework was initially published. In particular, we need to recognize that one of the main threat vector for identity theft is now malware. An identity trust framework can no longer ignore the potential of a man-in-the browser attacks (Trojans, key-loggers, worms, etc). Knowing whether the end user has any end-point protection (and maybe encouraging websites to introduce out-of-band messages into high assurance identity transactions when such protection is lacking) could be of consideration.
Trusting the transaction: from activity to security streams
Believing that the OP can provide strong identity assurance by simply checking credentials and abandoning the user at the RP front door is a dangerous over-simplification. Because modern exploits often let the user authenticate to commit fraud further down the session, it is important to enable OPs to leverage the knowledge of the end-user and her transaction patterns to identify high-risk conditions. Since we cannot assume the existence of adequate desktop protection (Internet security that exclusively relies on the presence of a client on the user desktop is no more than an academic exercise), high assurance federation models need to enable the use of fraud engines techniques across RPs (most logically, run at the OP although it could be a separate).
The ability to create an effective user risk profile across transactions is what has made the credit card networks work. High assurance identity networks are going to need an equivalent (think VISA of identity). An interesting idea could to leverage the concept of activity stream as a real-time fraud detection primitive.
A security stream back to the OP (under complete user consent and strict privacy protection) would allow RPs to feed transactional information back to the OP, allowing it to build a complete risk profile of the user across her Internet activities (fraud detection is often based on clustering techniques that measure abnormal deviation from normal behavior).
Even without a risk-engine running at the OP, a security activity stream could have tremendous security value if used as a simple identity alert system to notify the user of all ongoing transactions. In high risk cases, the activity stream could trigger an out-of-band consent for the transaction (think of Visa calling you to confirm and authorize a suspicious transaction); it is interesting to think that the social concept of activity stream that is today missing from OpenID (not from Facebook Connect) could actually be used to drive better identity theft protection. With such transactional feedback loop, a security minded OP would be able return a transaction score and possibly a liability guarantee based on the user risk and behavioral profile built over time.
Incidentally, interesting new OP business models could emerge (VISA-like: "I will take a cut of the transaction", Credit-Bureau-like: "I will charge you for the score", Insurance-like: "I will take the liability risk").
Ensuring trust across these three dimensions (the organization, the website and the user) is non-trivial. Yet, it is critical to enable consumers worldwide to engage into shared identity interactions with peace of mind across the Internet. Very much like PCI vendors emerged from the existence of a commercial PCI standard, one would hope that Identity trust assurance services could emerge as well since security companies need economic drivers to build great services. One of the key challenges of the standard will be to strike a balance between where to set the security bar to permit a high level of automation for accreditation. Such balance is always hard to strike, but it is also what makes the challenge worthwhile.
[Editor’s Note: Nico Popp is the VP of Innovation and Identity Management at Verisign and former CTO at RealNames. He maintains the Blue Ocean blog at Verisign. –mrh]