It does no good whatsoever to require every user of a grocery-list app to have a Yubikey to verify their identity. It might not even make sense to have users login at all.
The balance between usability and security must be consonant with the costs of implementation.
There could be (and probably are) entire books written about how to define what "an acceptable level" means... but that is the same point you are getting at - security is not a guaranteed lockdown of your assets, it is self-defined sufficient deterrence to attack. Sometimes that means light security, sometime that means heavy... but it is up to you to make those decisions.
Build yourself a threat model. It's only you that can decide this.
What may not be covered, and something which often causes problems with a lot of systems, is your threat model, may not be the same as that of your customers and, depending on what you're selling, you may not be able to know your customers threat model in advance.
To provide a couple of examples. If you provide server hosting, and a crypto exchange starts using your service, suddenly you may attract a load of attention from high-end attackers looking to compromise your systems as a means to get at other peoples.
or if you provide something like a consumer photo sharing/storage system, if "celebrities" start using it, suddenly you can find that people with a lot of time and interest start targeting you.
The tricky part is, commercially, do you have the resources to secure to the level required by the most sensitive customer...
Or as I often say "no one ever says, 'wow, that was a great login experience', they just want to get to the features behind that experience (hopefully securely behind it)".
People can make the right choice about security when they know what they are securing. Instead, online we are inundated with so much unnecessary security we start to have a prior biased towards: this isn't important. Back to the physical door analogy, when there are locks on every door even doors that really don't need them, people start propping open doors.
Sure, banks must enforce good security, but a lot of other things should not care.
Like online stores for example. As long as I am paying paypal, the only info my account has is email, address and recent orders. One can google my address in about 20 minites anyway, and this means I really don't care about account security. If there was an option to login without password using email alone, I'd probably pick it.
Security people always want to "set policy" "educate on practices" and "enforce". You've already lost the battle.
PROVIDE SOLUTIONS. Why recommend all this "policy" when what you need to do is provide, at a minimum, a reference implementation. If you get called in as part of security architecture, PROVIDE A SOLUTION.
Because if you don't the devs will do the absolute minimum, and likely will have backdoors galore, especially as your policies impose real restrictions on their systems support quality of life, ability to respond to production issues, and iterate to produce features.
The other persistent issue with security is that it is anathema to automation, and therefore efficiency. So dovetailing with providing a solution, these practices for 2FA and SSO (which invariably involves horrible popup UIs and other hacky things) will block, say, automated backups, auditing, monitoring, etc that also require access. So be ready with those.
>PROVIDE SOLUTIONS.
Here you go: https://qubes-os.org
Ideally technical security implementation should be seen as a function of the development/DevOps teams. You can have security teams to provide specific advice but the work of designing and implementing controls is best done within the team managing the system.
E.g., in Azure, in theory, we should PoLP the access controls. But Azure's tutorials and guides often recommend using Contributor (Azure role that entails access to almost everything, except granting more access), and which permissions an API call requires is, AFAICT, undocumented. And sometimes, the error doesn't tell you.¹
I want to allow SSH into systems. Copying keys about the landscape is one employee departure away from having keys on systems that don't need to be there. The last time I set up LDAP … I had to learn about object classes, and some sort of object-oriented tree database when all I want is a list of users & perms. (I understand LDAP's design better now, and I even like it, but it the onboarding is braindeath.)
There are any number of k8s dashboards that would give my coworkers better vis … and basically none that have an auth story.
The examples are endless.
¹heck, sometimes the error isn't even grammatically correct English.
For many platforms/services there will be security best practice sections on their sites and that's a starting point, but then as you mention even their tutorials often don't follow good practice.
The challenge for people writing standards docs is similar, new things come along all the time. How much time is availble to be dedicated to writing detailed guidance.
To give one example, the CIS benchmarks that a lot of orgs use to harden their environments are written almost purely by volunteers, so keeping them updated is a tricky game.
The 3rd edition is expansive (1000 pages, plenty of references) but readable. Free PDFs of previous editions are available at that link.
https://news.ycombinator.com/from?site=github.com/veeral-patel
I've been in the industry for 20+ years now and I can see things in that repo. that were old when I started :)
The technologies change, and the implementations change, but concepts like "defence in depth" don't.
:)