CERN accelerates towards usable security with new password policy

CERN is a European organization that hosts scientific research and labs for experiments, like the Large Hadron Collider. Their network connects the scientists and staff needed to support these research efforts. Despite being based in Switzerland CERN recently announced changes to more closely follow guidance from the US NIST SP 800 63B standard on user passwords in their environment.

These changes included removing password character complexity requirements and establishing a minimum password length of 15 characters. This latter measure is typically adopted to eliminate the more often guessed short, common passwords and encourage the use of longer passphrases.

With password character complexity requirements no longer in place to encourage difficult-to-guess passwords CERN will instead rely on two blacklists of forbidden choices. The first is composed of simple passwords (like ‘123456’ and ‘CERN2025’), and the second contains “burnt” passwords. These so-called burnt passwords are publicly known by at least some password hackers. CERN learns of these by using the HaveIBeenPwned database and other repositories of passwords publicly exposed through data breaches.

CERN had already stopped forcing regular password changes with an annual expiration policy back in 2020. At that same time they’d implemented an adaptive password policy similar to the one the University of Pennsylvania recently adopted. Why that policy has now been simplified further to just a minimum password length isn’t discussed, but it may be to further reduce user confusion about how to create a compliant password. CERN was finalizing their deployment of Two-Factor Authentication (2FA) to users last year, so the security added with that change may have also reduced the need for a strict password policy.

Link to announcement: https://home.cern/news/news/computing/computer-security-password-evolutions

Paper Highlights: Investigating the Password Policy Practices of Website Administrators

This paper, "Investigating the Password Policy Practices of Website Administrators", was presented at the 2023 IEEE Symposium on Security and Privacy conference. But I ran across it today and thought it provided some helpful insight into why people developing or maintaining web applications chose certain password policies. The research team interviewed a small sample of 11 US-based professionals who had experience setting or managing website password policies in order to learn not just what decisions they made, but why. These weren't necessarily dedicated security team members, but more likely developers or system administrators.

A few highlights from my read:

Password composition restrictions (e.g. what characters or what length can be used) were often a result of a compatibility requirements with existing systems at the organization. Some of these restrictions affected common symbols (e.g. "&" and "?"), but others were probably extended ASCII or Unicode characters.

One organization was still limiting passwords to 16 maximum characters because of the contentious logic that 'limiting the length was necessary because users often forgot long passwords'. A couple others didn't place any limits on maximum length.

7 of the 11 respondents said they were still enforcing password expiration despite some industry guidance starting to discourage this practice. They seemed to think this provided needed protection against account takeover (ATO) from leaked or shared passwords. Those who didn't force expiration referred to their concerns that regular changes caused more user frustration and felt their systems were secure enough to withstand password attacks.

About half the participants mentioned looking either at industry standards (like NIST's 800-63B) or the practices of other large Internet sites (like Facebook or Google) for guidance on forming their own password policies. A few cited legal or industry compliance pressure forcing certain settings.

There are other interesting disclosures, like whether these organizations blocked certain passwords (e.g. blacklists) and how they decided what passwords to block. But I'd also like to hear from those of you who have been involved in this process yourselves. What steered some of your decision making?

Paper link: https://www.computer.org/csdl/proceedings-article/sp/2023/933600b437/1OXGTWy2ktq 

FTC orders CafePress not to store security question answers in plaintext following breach

CafePress is a business that specializes in allowing users to create custom merchandise, like graphic t-shirts, and use their online store to handle sales and fulfillment. After discovering they had suffered a breach in early 2019 the company quietly required users to change passwords while claiming this was due to a password policy change. However, a few months later it became apparent the 23 million record user database containing both buyer and seller customer accounts had been compromised when it was posted online for sale by the criminals, and CafePress was forced to admit they had been hacked.

The US Federal Trade Commission (FTC) got involved as part of their mission to protect consumer privacy and filed an official complaint that highlighted the shortcomings of CafePress. This started a process that would determine what security improvements, ongoing assessments, and fines would be required of CafePress. They issued their final Decision report (PDF) in June of 2022.

Among the many faults outlined in the initial complaint were details of how CafePress didn’t take “reasonable security measures” to prevent the exposure of sensitive user information. The breach had exposed unsalted SHA-1 hashed passwords, security questions & answers, shipping addresses, and US Social Security Numbers (SSNs) for some sellers.

The FTC highlighted the fact that while CafePress had required customer password changes following the breach they didn’t force changes to security question answers. And these security questions were used for account recovery. It appears that after requesting a password reset the users were prompted with their security question and allowed to change their password directly after answering it correctly, without any email verification needed. So the original attackers, or anyone else that had obtained the stolen data, could perform account takeover (ATO) by plugging in leaked email addresses and security question answers.

Related to this problem, the FTC highlighted that storing these security question answers in plaintext was not adequate protection. But if CafePress could hash passwords -- albeit poorly -- then why were the security question answers stored in plaintext? The short answer is that most information in databases is stored in plaintext by default. Unless someone involved with the software development process identifies that this practice is either too risky or that it fails to comply with laws/industry standards then that data is likely to stay unprotected.

The slightly longer answer is that some of the systems that manage security questions do expect to have plaintext access to their answers. Unlike passwords that tend to require exact matches, answers to security questions are sometimes given more leeway as long as they are close enough to the expected answer. For example, the question “what was your first address” might be answered “123 First Street” or “123 1st St” depending on how the user is recalling their address. Some systems even accommodate different character capitalizations “123 first street”, typos like “123 Frist Street”, or missing words “123 First”.

There are also situations when the same security questions used for online access are also asked by customer service representatives talking to customers over the phone or in person, possibly requiring these personnel to see the customer’s answer to check it for correctness.

So when hashing answers is not possible, what is the alternative? These answers could be encrypted before storage. Encrypting these records (along with proper key management and access controls) could allow the answers to be decrypted and checked when necessary without exposing them to any attacker with read access to the database.

Interestingly, the FTC didn’t actually recommend that CafePress encrypt their security question answers, but ordered them to get rid of the questions altogether. They wrote that multi-factor authentication (MFA) alternatives should replace this functionality. I’d argue this directive doesn’t clearly address the issue of account recovery, because that can still be a problem even with MFA, but it does eliminate reliance on security questions as the sole gatekeeper of the recovery process.

If you are going to continue to rely on security questions it seems like you should avoid some potential legal and financial trouble by protecting their answers with encryption, as well as force users to change them if you ever suspect the data has been compromised. Then you just have to deal with all the other problems of security questions.

Thoughts on new authentication guidance in OWASP Application Security Verification Standard (ASVS) v4.0

[This content was originally posted in a series of tweets, but it also made sense to share it here.]

OWASP released v4.0 of the Application Security Verification Standard (ASVS) in March, listing security practices for organizations to design, code, and test apps against. There were substantial content changes in the authentication section, so after reviewing it I wanted to tell you what I thought about the changes.

The authors state in the V2 Authentication Verification Requirements section that their goal is bringing this standard closer in line with significant authentication changes published by NIST in the SP 800-63 Digital Identity Guidelines update that came out after ASVS v3.0.

There are 57 requirements in section 2 for ASVS version 4.0, compared to 26 in the same section of 3.0, which more than doubles the number of previous requirements.  Around 9 requirements are seemingly removed in 4.0. So let's go through some of the notable changes to the standard.

Requirement 2.1.1 establishes a minimum password length of 12 characters for users, which is one big divergence from NIST’s minimum of 8 chars. While this is justifiable for security, I do think it will cause protests, especially if dealing with third-party or legacy apps that have hard-coded lower length settings.

2.1.4 asks for support of Unicode characters in passwords. Another good change, and while I don’t have statistics I suspect a large number of Internet apps can’t meet this requirement today. Many of them are still struggling just to allow symbols (see @PWTooStrong).

2.1.6 talks about verifying the old password to select a new password, but removes text from older standard about new password confirmation. I suspect that with the addition of 2.1.12 saying to add a password field unmasking option that OWASP made this change for usability (see Why the Confirm Password Field Must Die).

Requirement 2.1.7 expands on previous guidance to prevent use of common/weak passwords to specifically recommend use of a 1,000 - 10,000 entry blacklist, either maintained locally or transmitted securely using a third party like @haveibeenpwned or Azure AD Password Protection.

2.1.9 says to eliminate any password complexity policy requirements or restrictions. This places responsibility on blacklists and minimum password lengths to prevent bad password choices. It's a very contentious change for orgs who've used complexity policies for decades.

Speaking more generally, the migration away from password complexity policies to blacklists is a major shift that needs additional research. If you implement password blacklisting within your org please find a way to share your lessons learned, anonymously if needed, so we can all benefit.

New requirement 2.1.8 says to provide a password strength meter to guide users towards choosing stronger passwords or passphrases. However, not all meters are created equally, so I recommend taking the time to select a good one (see On the Accuracy of Password Strength Meters)

2.1.10 instructs the removal of password expiration policies, which has gained support in recent years. But this goes hand-in-hand with requirement 2.2.1 to implement controls to combat password attacks and reduce the chances of a password compromise leading to account takeover.

Requirement 2.2.2 discourages relying on ‘weak authenticators’ like SMS and email. 2.2.4 and 2.2.7 advocate prioritizing reliance on MFA options less likely to be compromised, like OTP tokencodes, U2F security keys, or client-side certificates.

2.2.3 says to securely notify users following any changes to their credentials, emails, addresses, or if new logins to their accounts occur from a riskier/previously unknown location. The use of push notifications is preferred to better direct user attention to these events.

Requirements in 2.4 expand OWASP guidance on password storage to include aspects like salt randomness and length. But they don’t mention Argon2 or scrypt, instead offering work factor advice for bcrypt and an oddly inflated PBKDF2 iteration count of 100,000 (NIST is satisfied with 10,000).

They do discuss Argon2 and scrypt in the OWASP Password Storage Cheat Sheet, so it’s likely these documents just need to be aligned better. The PBKDF2 iteration count discrepancy may just be a typo.

2.5.2 now calls for the elimination of security questions rather than just making sure the ones in use are ‘good’. Many industries, especially US banking, still rely on these for secondary authentication. But with stronger MFA options OWASP thinks they can be phased out.

Sections 2.6 and 2.7 are brand new and describe recommended security elements for Transaction Authorization Numbers (TANs) and out-of-band (OOB) authenticators. These tie back directly to NIST guidance for designing or implementing these solutions.

Section 2.8 likewise gives guidance on one-time password (OTP) authenticator use, with section 2.9 discussing cryptographic security key authenticators (FIDO U2F). You may not develop these functions in-house but should validate that your vendor has done so properly.

Section 2.10 adds further guidance on storage of passwords or API keys used by application code, including not to “rely on unchanging passwords”. I think a better word would be “unchangeable” so they should be updated as needed but not on a set schedule.

So what was removed in section 2 of the new 4.0 ASVS? Some practices in the old standard may have been considered too basic, such as ‘require authentication for anything non-public’, ‘enforce authentication on the server’, and ‘fail securely to a default deny.’

The old 2.2 is removed which stated “Verify that forms containing credentials are not filled in by the application.” This could be understood to include password managers autofilling credentials, rather than just browsers, which OWASP may have decided to no longer discourage.

One absence is the old 2.23 practice of making sure account lockout due to login failures was flagged separately from administrative account disabling. It was a logical recommendation, so I don’t know if it was also considered too basic or whether it's removal is an oversight.

Also removed is the 2.28 guidance to ‘make sure all authentication challenge responses take the same time.’ Intended to prevent things like timing attacks that leak credential info, this may have been thought to be too difficult to implement consistently for the value it provided.

Old practice 2.32 wanted you to make sure administrative interfaces weren’t accessible to untrusted networks. This may have been removed because it becomes more difficult in a cloud-hosted world, or maybe because some apps use the same login interface for admins and normal users.  [project lead Andrew van der Stock commented "We did that because "where" is not as important as "who". The idea of Fortress Admin is laughable and always has been. Let's move beyond layer 3 restrictions."]

Finally, gone is the old 2.18 requirement to verify that username enumeration isn’t possible in login or account recovery functions. While preventing account enumeration is good, it tends to provide little value at the expense of the user experience.

I created a Google spreadsheet to compare the section 2 authentication requirements of the ASVS 4.0 and 3.0 side by side. You can access it here: https://docs.google.com/spreadsheets/d/1UbOsbgv4WsmuVuL8M3NoCRD7UQKAw7vl6BLYaLk-EtI/

Wrapping up this review, I want to thank the OWASP organizers and volunteers who developed this standard and made the tough decisions about what practices to include/exclude. Project leads include @vanderaj @JoshCGrossman @dcuthbert @m8urnett & @manicode.

Banks fail to meet FFIEC multi-factor authentication requirements

According to a recent study, only 4% of financial institutions are actually meeting the FFIEC requirements for multi-factor authentication. Many of the others are relying on challenge questions or other risk-based authentication approaches that ultimately only provide single-factor security.

I’ve shared more on my analysis of this problem and its implications in this Security PS blog entry.