Telling the Truth About Security
In James Morrow’s wonderful novella City of Truth, people in the city of Veritas have been conditioned to tell the full, unvarnished truth in every situation. They drive a “Renaut Adequate,” send their children to Camp Ditch-the-Kids, and always pay their bar tabs. The warning on cigarette packs reads:
WARNING: The Surgeon General’s crusade against this product may distract you from the myriad ways your govenrment fails to protect your health.
I think about this often in the context of security messaging for people who use the software we produce. Real warnings on cigarette packs are one of the few examples of honesty in packaging we have today, but that’s not good enough for Veritas.
What is the right amount of honesty for messaging to the end user? Clearly dishonesty is bad, but so is a dissertation-length discussion of encryption parameters. If our assessment of real-world threats creates an excessive paranoia in end users, convincing them that APT crews are coming for their kids, then our “honesty” becomes dishonest. I would assert, without a lot of evidence to back me up, that the right amount of honesty is that which enables the user to make good decisions about their behaviors, within the application’s threat model.
That “within the application’s threat model” hanging off the end is what allows me to say that things which are true but distracting should not be included in security messaging.
Want to Start an Email Service?
I was thinking about this in the context of some tweets from Nathan Buuck. He wrote:
I’m curious to sign up for an email service as a new user and see what they warn you about as a presumably-new email user. I imagine most services - especially those that want to keep you for ad revenue - don’t give you a full explainer for fear of scaring you off. Ethically, though, every new email user should be given a briefing about how:
- you can’t trust anything you see in email at face value
- your digital identity will grow to depend existentially on this email account
And the litany of other concerns.
Failing to prepare new users by not illustrating to them the potential scope of the risks we know they may face is just negligence.
Well, you can tell that these are good questions because there’s a ton to unpack here.1
One thing that’s interesting to me here is the difference between Nathan’s two bullet points. To whatever extent the first point is correct, it means that the security of the email service has failed. Whereas to whatever extent the second point is correct, it represents a failure of security that’s the fault of nearly everyone but the email service. I’ll explain more below.
Because the risk of compromise through email is so high, we’re at a real risk of erring on the side of over-explanation, especially if we try to explain it all at once, but even if we don’t. If we consider only email, only security, and only the items which all users need to understand,2 that’s at least:
- Basic scam detection — if it looks too good to be true…
- Phishing, credential harvesting
- Spearphishing, and why not to wire money to your boss
- Basic identity/password hygiene, multi-factor identification
- Physical security — don’t go meet strangers you’ve corresponded with
- Dark patterns in advertising
- Common user misbehavior, such as chain emails
This is a lot for a user who signs up in order to stay in touch with their kids to take in! But the security scope creep will start immediately, partially because of the actions of criminals, but partially also because of the decisions of the security community.
When Should We Explain This?
I don’t think that during sign-up is the correct place for explaining what you can and can’t trust in an email. The right time to explain phishing is 30 seconds before you are phished. It’s not always clear when this will happen, but I think it’s fair to say that during the sign-up process is perhaps the least likely time that the user will be phished in email. Putting this in onboarding feels like a legal cop-out to me: “We told them; what else could we do?”
But there is a responsibility to explain this stuff at some point! The old model of an email client blindly displaying whatever you send it doesn’t reflect the reality of a hostile world. Today, for example, Gmail will tell you if someone you correspond with suddenly begins emailing you from a new address. Maybe it’s someone trying to spearphish you, or maybe it’s just JIRA. At any rate, such emails deserve extra scrutiny! Putting this on the suspicious email itself is a much better solution than putting an explanation on the screen during signup.
What Do We Ask Users to Trust?
Since Nathan’s question asks what email services display to new users, I’d like to reframe this point in terms of the service provider: What does an email service ask its users to (dis)trust?
When Nathan says, “Failing to prepare new users by not illustrating to them the potential scope of the risks we know they may face is just negligence,” he’s not wrong, but we have to be really careful about sentences like this, because they include risks both inside and outside the threat model of the service. For example, training is the wrong approach to the problem of “attachments in email might harm your computer.” The right approach is to make clicking on attachments in email safe, not relying on end users to make good security decisions 100% of the time.
Indeed, nearly every company has people in HR who must click on PDF files attached to emails from random strangers on the Internet as an essential function of their job! We can’t tell people “Hey, don’t do your job; it’s not safe.” We have to give them an email client where this mundane action is safe.
Example: Phishing Simulations
Implicit in the notion of what an email service should disclose to / train users about is the presumption that we even know what to tell them at all. Unfortunately, we cannot take this as a given.
Phishing simulation has become fairly big business these days. It makes sense, as phishing is often the attacker’s first foothold into an enterprise. It’s important to have clear goals for the phishing simulation, since, like a red team test, the attackers always win.3 If you conduct a phishing simulation and nobody is fooled, then your simulation was unlike an actual phish, which will probably result in users being deceived.
Therefore, if the purpose of your simulation is to determine if anyone will fall for a good phish, I can save you the money: Yes, they will. If you fire everyone who falls for the first phish, like some perverse form of stack ranking, then you’ll have more people who are taken in by the second. However, phishing simulations can be useful if you have constructive goals in mind.
“Lets be honest with each other. Phishing simulations aren’t just about training. They are also popular because they produce a metric (e.g. ‘Last week 60% of people fell for our phish, this week only 35% fell for it’). It appears really positive and encouraging, since it appears to show that something is being achieved, but unless you’re careful you might just end up wasting time and effort.”
“The Trouble with Phishing,” Kate R, National Cyber Security Centre, UK
Here are some constructive goals we might set for a phishing simulation:
- Train users to send phishes to the SOC, giving the blue team a heads up on what’s in the field
- Train users how to respond if they click on a phish and enter their credentials. If we can train users to call the SOC immediately instead of keeping quiet and hoping nobody notices, that’s a big win!
- Increase the level of end user skepticism about what’s in email. I feel a little weird about including this, since, as I said, the phishers will always win. But we can certainly make them win less if we treat this as proactive training, not punishment for people who click on a phish.
Unfortunately, if you look at the promotion materials for phishing simulation software, they’re all graphs of “number of employees caught,” which suggests that security management is pursuing metrics instead of better security. This attitude is noticed by employees, who start to regard the security team as people who are out to “catch” them, because they are. If we train employees to distrust the security team, we have badly failied at our job.
This is a good example of why technical or legal controls are not enough to protect users of an email service. The training we give must be informed by human-centered, compassionate support for people. Running it by the corporate equivalent of an IRB would not be a bad idea. And we must repeat this introspection with every control on our list.
Your Identity Source of Truth
To Nathan’s second point, “your digital identity will grow to depend existentially on this email account,” well, that’s correct, but it’s interesting to note that this is not because of some technical factor which makes email the unique thing that is the only possible source of truth for identity, but rather because, in a nutshell, the entire industry has decided that email — and by “email” we mean Gmail, mostly — does security well, and therefore the source of truth for identity should be Gmail.
As evidence for this, consider that it was pretty common to use SMS as a source of truth for identity until it became obvious that SMS could not be trusted. Services are moving away from SMS, and Gmail is the last thing standing. Even sites supporting MFA — a minority! — often use email as a portion of their password reset process.
This would be less of a security choke point if people always made a unique email address for each web account they created, but almost nobody does this. So if your email is compromised, Nathan is correct, the bad actors get every account you have, including retirement accounts (your life savings, uninsured), potentially embarassing accounts, etc.
It will be interesting to see if WebAuthn, which doesn’t require an email address and appears to be secure, has similar uptake in the future. It will also be interesting to see if the uptake of less secure email services will result in people moving away from email as the identity source of truth. The Yahoo! breach did not push developers away, though, probably because the other options were all worse.
Do I Want to Start an Email Service? Hell, No!
One thing is very clear to me: When you start an email service, to whatever extent you’re successful and your service becomes popular, you are painting a giant target on your back and on the backs of your users. Your security team — meaning not just your blue team but also the people who work with human computer interaction in the security space — must be very good, the best people in the industry. If you don’t want to take that on, you have no business starting an email service.
One hip email service brags that they “engaged two separate, external security firms to review all our application security.” That simply doesn’t cut it; it’s laughably inadequate for the task at hand. How would hiring an external team protect users from a phishing attack? The problem of email security is just much larger than technical controls.
Leaving Veritas
In City of Truth, the protagonist, Jack Sperry, falls in with a group of (literally) underground rebels who sometimes lie. He meets his roommate, Ira Temple.
Ira, I learned, was a typical dissembler-in-training. He hated Veritas. He had to get out. Anything, he argued, even dishonesty, was superior to what he called his native city’s confusion of the empirical with the true.
I hope by now it’s clear why I don’t think it’s sufficient to “tell the truth” about security; we must have a more nuanced conversation which is not just at sign-up and continues throughout the user’s use of the service. It must always be informed by the amount of information a typical user can take in at any one point in time. “Confusion of the empirical with the true” describes a lot of mistakes we make in software.
Comments on this post elsewhere
I do think advertising is a red herring here. The two points in the tweets, while really worth discussing, don’t change with or without ads. Indeed, a very well-known email service with ads also does a better than average job identifying what you can or cannot trust in email. I think ads are orthogonal to these questions.↩︎
By “only email” I mean to exclude things like encryption which are security-related but not directly relevant to basic email use. By “only security” I’m excluding things like how to compose and send emails, which are really important for new users, but out of scope for this article. By “only the items which all users need to understand” I’m excluding things such as nation state attacks, which are security related and relevant to email, but not something that the average user needs to think about.↩︎
Nobody, and I mean nobody, is immune to phishing. The organizer of the DEFCON Social Engineering Village recently shared his story of getting phished.↩︎