Edit: e.g. How do you get threat intelligence, compliance assessment, and a shareable security architecture to assure customers/stakeholders into a backlog?
- Most customers write design docs/tickets. This helps because we can suggest alternative approaches early. It’s rare that we tell people that an idea is fundamentally flawed, but it’s common that we suggest a different implementation approach or additional safeguards.
- We review individual PRs. Here, we're looking for concrete individual appsec concerns (or infrasec if it’s a Terraform PR for example). Not the right place to have overarching product design conversations.
One concern is that there are just too many PRs for a fledgling security team to review every single one. For this we rely on 2 mechanisms:
1) IC opt-in (they tag us on GitHub via @CLIENT/security when they would like a set of eyes on something or if a PR is part of a product change we’ve been involved in)
2) Automated heuristics. These have a hair trigger: lots of false positives. It’s just stuff like “are you adding middleware?”, “are you adding an endpoint?”, et cetera. That’s OK, because we don’t make it a gating part of CI: its job is to turn 100 PRs for us to look at into 10 PRs for us to skim and make a decision if it’s the 1 PR that deserves extensive scrutiny.
So, overall: stop thinking of yourself as a gatekeeper and start thinking of yourself as an advisor. Our job is to make sure you understand what you’re doing and what your alternatives are. Security is never the only requirement: you have a business to run.
Individual evaluations don't capture risk you get from just having a lot of dependencies. "What happens if this gets compromised" is something we consider individually, but it becomes more actionable when you look at your dependencies globally. Examples: someone loses their Chrome Store credentials, or someone accidentally pushes AWS keys to GitHub, or someone loses their npm creds in an eslint-scope style problem. Now the dependency itself is malicious. If it's a library, that means code execution on your developer laptops, CI systems, and possibly customer devices.
Individual dependency requests are not the right place to push back on introducing a dependency. You need a second speed to address those by teaming up with e.g. the VPEng and usually someone in PeopleOps/HR/Sales/Marketing (for things like calendar mgmt, sales tools, ad-hoc CRMs) to periodically pare that list down.
Cloud security architect at a financial services firm here (supporting thousands of devs). We're using the usual pull request review process between dev/ops teams and our risk management team (which is composed of various subject matter experts) with an SLA. We're also hard gating ("you shall not pass") using Sentinel rules for Terraform Enterprise and service control policies in AWS accounts.
I highly recommend you put a training program together that incorporates a feedback loop from the above process. Before devs and ops folks start, they should be gated until they go through your training process. That should minimize the exceptions you'll need to catch in the review process, but when those exceptions occur, that should be noted for training improvement.
Really, above all else it's just hiring or contracting (ideally hiring) good security people. Good appsec people to review code and look for vulns, people who just do education and training, people who just do penetration testing, people who just do security architecture and tooling. And of course a good security operations and incident response team to deal with the day-to-day "frontline" defense, investigation, and response. If they are full-time employees who are integrated with the company culture and are experienced enough with agile development, then odds are they're not going to set up more red tape than is necessary.
If security is mostly offloaded to devs or operations and no independent security org exists at your company, you're in trouble.
Any vulnerability, misconfiguration, or exposure could be exploited by many different kinds of entities for many reasons. A company with a good security department will have a dedicated threat intelligence and research team to consider specific risks and adversaries, which they will share with adjacent security teams; that information probably isn't relevant to most people just working on a product. But sure, some of it may reach a product owner if there's a particular kind of risk inherent to it (like if the product is offering secure end-to-end encrypted communications as a selling point or if they're in a sector that's been heavily targeted by state actors).
Every organization and product will differ a lot, but for any company with a not-insignificant presence in an industry, state actors (APTs) are pretty much always going to be a risk. The odds of being hit by an APT are pretty low for most small companies, but the risk is always there.
In terms of what's most likely to regularly affect the average company, financially motivated cybercrime has pretty much always been and remains to be the biggest issue. Much like regular daily crime, I suppose. It's people just looking to make a buck. Commodity malware like ransomware and banking info stealers, vuln scanning bots that pop a shell on their web servers and sell access, phishing emails spreading malware or impersonating executives and requesting wire transfers are probably still the main risk for most companies out there, of all sizes. This is probably the biggest problem for most companies whether they're entirely tech/software-focused or just a retail outlet or restaurant.
Sometimes you'll get more organized and sophisticated cybercrime actors targeting specific organizations in an APT-style manner, but it's still almost always using those tried-and-true basic tactics. And believe it or not, many state-sponsored APTs themselves often go with generic tactics, too. Most influential countries' intelligence services do seem to have at least one "A Team" conducting cyberwarfare/espionage, but some also seem to have a lot of grunts just looking for low-hanging fruit, and they're usually not much harder to detect or stop than more typical script kiddies you might otherwise encounter. Often the "A Teams" will intentionally stick to the most generic stuff they can find, too, to avoid tipping their hand that they're smarter than they seem and to hide the fact that they have access to non-public tools and exploits (both to prevent their "weapons" from being burned and to not seem like they're an APT). The most sophisticated groups will only pull out the big guns if all of the generic techniques fail.
I'm not really well-versed in industrial espionage and competitors reverse engineering software, but I think it's not really something most companies need to care about. If your competitive advantage is actually at risk just because someone reverse engineered one of your binaries, there's probably something very wrong with your business model. The companies who need to care about that kind of stuff are usually already well-aware.
The obvious threat scenarios above are examples of things product managers typically don't think about, but developers assume someone must have understood and decided. I'm trying to ascertain if this gap is a broad problem, or just a dark pattern in organizations that could be by design:)
This is where large companies might trip and fall, if they expect their existing security analysts/teams to do this (or, as you say, if they try to offload it onto developers or ops people, even ones who are more interested in security). I don't like how this sounds, what I am about to type, but the majority of security people -- particularly the ones at the level where they are hired to run QA-type audits -- do not know how to review code. It really does make me uncomfortable to say this, but if that's the rock, the hard place is how many times I've had to tell someone they really need to learn to code if they want to "work in security."
So when you say, "good security people," what you mean is people who can read and write code, and also have enough experience to know when that code can be used in a way that the original author did not intend to. It's almost as if there's a misconception that being able to read and write code is what you do if you're a dev or devops. This mindset is odd to me in so many ways. It's like saying, "I don't need to know how to use a screwdriver or a hammer because I'm not a carpenter." Even better, someone who understands what it feels like to be under some pressure to roll out changes where there might not have been a lot of time to consider malicious behavior. Someone who has your back.
Just yesterday or the day before, I had this inexplicable crisis where I was concerned about the value I am providing. It's gone now. Thanks!
For others, their appsec team(s) is/are constantly building security libraries, frameworks, and tooling, and scanning code for security issues with software and manual review on a daily basis.
Information security is still just a checkbox for a lot of companies. This is gradually changing with more and more breaches in the news every few days and executives who are finally starting to appreciate that the consequences of a breach can be very bad, but it's still pretty common. I really don't think many companies have solid appsec teams that are doing the things you and I would hope they would do, and I agree that probably a scarily high percentage of "application security analysts/engineers" do not and cannot review code effectively.
You are absolutely right that knowing how to write and read code are crucial skills for many aspects of infosec and that a lot of people neglect that, and it's disheartening that there are many companies who don't really have people like that on their security teams - even companies that claim to have an appsec program. But application security is also only one aspect of a solid information security program, and some other aspects do not necessarily require development knowledge beyond the basics.
From what I know and have experienced to a small extent, FAANG (and others in/near that tier) invest a ton into application security and really are doing it right, at least. (Or are at least doing it way better than 99% of other companies out there.)
We are not allowed to release until the security scans are done and reviewed with the app security team. We release about once a quarter because of this. We do not practice continuous delivery and doing agile is difficult. There's no gatekeeper, story grooming, product owners, nada. You do what app security says when they say to do it, we work on their schedule.
As a tangent, working for a large company has shown me that agile doesn't really work at this scale. There's too many third parties that you have no control over. And the business is always trying to respond to the market, so projects are changing all the time and people are always shifting roles. I just come in, put my commits on the branch, and pray that the machine has my back.
For example, we were preparing to run some world cup soccer promotions based on the teams that won. We literally had to wait until a soccer match finished in order to deliver our changes. How do you practice agile with that kind of restriction?
One thing we've found to be effective here: we start with a comprehensive audit (appsec, infrasec, corpsec, deployment security, external network footprint) that tells us about systemic problems you face. Then we can provide targeted training where you contextualize the problem for ICs.
I'm sure Ryan McGee (magoo) has ways to tie that in to quantitatively making people better at risk estimates, but we're not trying to get ICs at clients to do that (yet).