You’ll get much more bang for buck for the product’s security by working on the enterprise and infra side - deploying EDRs (somehow not on this list?), tightening up Gsuite sharing and email settings, Okta (I see SSO on the list), anti-phishing capability, guardduty and sechub, a pw manager, getting on IaC early to support ops and sec goals, and a lightweight SlackOps infra for security alerts. Somehow mostly none of this is mentioned.
The emphasis around the application seems misguided due to how (a) pragmatically it’s going to take much longer to get appsec controls and logs vs infra and enterprise controls/logs and (b) the vector in is usually Bob and Jodie in recruiting and HR getting phished, vs an appsec breach.
Also, it seems to break the golden rule of security enabling revenue with acceptable risk trade-offs and pragmatic controls. All the process in here doesn’t seem pragmatic. The controls in here seem useful overall but not where I’d go at first to secure a fast startup.
I am not a security engineer. But as a generalist software engineer I do not believe this. Every time I go looking in app code with an eye on security, I immediately find all kinds of vulnerabilities.
And yet, where I work, the security guys don't read or write a single line of app code. They do like you say, and focus almost entirely on the infra/business process side of things.
I find it really hard to respect these people when they are more worried about locking down what jira tickets I can view than they are actually securing our product.
Preventing breaches is an objective, but it's purely secondary to CYA. This is why the Security Strategy Policy exists and often looks so out of touch with reality:
1. An external consulting firm advised on that policy document.
2. Someone signed off on that policy document (CEO/CTO/COO/etc).
3. Everyone below that person ticked-off every single clause in that document.
4. Breach occurs.
5. Employees can't be blamed, they did everything that was asked of them
6. CEO/CTO/sign-off person cannot be blamed: they sought, paid-for, received and heeded professional advice.
7. The external firm cannot be blamed, as their advice is "in line with modern security guidelines."
In this sense, no one should be shifting focus from whatever the policy document says because then they can be blamed.
This risks that happen that lead into bad exploits for startups and the controls available:
- SaaS vendor is breached and doesn’t tell you for a week: okta allows you to cleanly kill perks for that SaaS, and impacted users
- dev downloads a database access package on vscode, puts in prod creds, the package is calling home bc it’s surprise malware: EDR catches it
- dev has prod passwords on their MacBook note sheets, or accessing prod from a home computer: process enforcement, password manager, secrets vault
- customer service employee is phished, pivots into customer data: IR plans and email security tools
- product m API is “move fast and break things” and full of vulns: except for the worst of them, get a WAF/cloud flare to cover and be careful about public endpoints beyond that.
The above secures a product with limited bandwidth and a small security team, if you’re thinking defense in depth (which you should be).
It’s important for sec engs to code, no doubt. Two things to note that I think you’re missing, putting aside issues if your sec team is doing Jira ticket focuses vs the above.
- code vulns are paired with exploit paths. If a WAF is there, and the code doesn’t pivot into prod DBs and so on, these can wait to remediate.
- Have you tried remediating code vulns or getting application logs for security, at a product-focused startup? Devs are the worst enemy here. Anything touching the code is “slowing the product” usually. Security is a political job and this is an easy way to burn all your social capital if you chase each of vuln.
In short, you’re thinking about vulns, but not threat models. Startup sec program is all about smart thread modeling.
Considering most of the major incidents over the last few years fall into these categories, it makes sense.
Will edit with commentary in a moment.
Edit: alright, fair criticism re business security controls given that the checklist's first section is specifically about business controls.
And the rest should be focusing on automating those outcomes rather than just what the outcomes are. I agree that manually creating DFDs is just going to slow teams down, though it's definitely a capability that more security-mature or regulation-burdened organizations will need. When you're this early though, better to harden IaCs based on secure reference architectures up front and lint the heck out of code than to encumber all of ones processes with manual controls.
Unless you have the unicorn dev who’s doing appsec also focusing largely on appsec early on is a losing battle politically, will get your 1x sec eng at the startup - who also has to do a ton of other stuff for the startup - to quit for a better job in 12 months, and so on.
Defense in depth and threat modeling needs to rule a security program for a startup and doing MSVP as your guideline would miss that nuance.
If you're working for a startup, we would greatly appreciate it if you could try implementing MVSP in your organization and let us know which controls were difficult or problematic to implement. These controls can potentially be considered for removal.
Thank you for your feedback! I've carefully read all the comments in this thread, and it seems there's a misconception that MVSP replaces or substitutes enterprise or compliance controls.
The goal of MVSP is to add the "S" in Minimum Viable Product (MVP). It establishes a minimum standard for a software artifact, like a SaaS app, to be considered for purchase by companies that use MVSP as an RFP baseline.
MVSP originated from the security contracts' language used by big corporations like Salesforce, Google, Dropbox, and others in the FAANG MANGA group.
If you believe that certain requirements could be removed, please feel free to join our mailing list and share your feedback for discussion: https://mvsp.groups.io/g/mvsp
Our next working group meeting in the US timezone is scheduled for September.
That’s the goal of running a security program for a startup, ultimately - get them to Series C without a serious breach, and then build out a security team to get them to IPO and through SOC2 etc.
So my concern is that if you have just one shot to standup a security program, and the goal is the above, building out a appsec program with the MSVP (which also skews into a bunch of process, beyond appsec) seems to veer far into stuff that’d be unproductive for startup security team to do.
Reason being is it hits what I said above in another thread - unless you have a dev knowledgeable and willing to do all the suggested appsec controls in the MSVP, that’s a sec eng asking for each of those controls. The odds of a clean devops pipeline even existing yet seems low, as to slot in SAST and some DAST approach to catch the vulns suggested by the MSVP. Or the budget for a pentest.
That’s a sec eng annoying startup devs for constant code changes for code that works and is getting revenue (bc this is how the 101 startup dev usually sees these requests).
That’s also a sec eng who is usually doing semi-GRC, supporting sales processes, doing all the above enterprise sec 101 stuff, and potentially ops/IT support, as there’s usually only 1 or 2 of these hires early on.
So, makes me think that the MSVP as written misses the pragmatic reality of what works for startup sec programs, although I’d agree it’d work for a product. But to secure the product given realities for a startup dev, WAF for the product endpoints helps deal with the worst appsec problems and then use defense in depth on the enterprise side. Phishing the customer service rep, SaaS breach, or sketchy packages are very commonly what gets startups.
It is about the overall attitude. One cannot expect people to think about secure code, when they don't follow basic principles outside of the product itself.
> for enterprise-ready products and services
Are we seeing something different?
Can you tell us more about that? What kind of capabilities can protect you against social engineering?
While I agree Okta is the most feature complete why only Okta? Are there any other products that can replace Okta?
Jumpcloud got breached badly recently. Okta has as well but their breach wasn’t the same.
Generally with IDPs, worth going with a gold standard that works well. Okta is that
Okta is definitely the most complete, but plenty of other options.
Whoever wrote this must be so irrationally out of touch with the startup space (or thinking of older billion dollar unicorn startup) to think that an MVP needs to do any of the above. I wouldn't care about GDPR until I have a somewhat strong EU userbase. Try to respect the spirit sure, but it's not like it will be enforced on a small business within it's first few years of existence.
Localization for an MVP is even more out of touch. Make an English US-centric version first (I'm not even American) put it out there and work on localization once you've had some success.
Focus on value, everything else is fluff. When you have a few paying customers and that your product is now mostly proven then it's time to make it better.
And really, most of them don't provide security, they're a checklist. Checklists don't provide security, they provide (sort of) accountability.
This seems ethically wrong -
1. Even if your user base is not “strong” the privacy of the users you do have still matters
and
2. It should be about protecting them not just calculating enforcement risk like a Ford Pinto engineer.
Just off the top of my head, things that aren't even universally seen by practitioners as good things, let alone things everyone does as a "minimal" baseline:
* On request, enable your customers or their delegates to test the security of your application
* Implement role-specific security training for your personnel that is relevant to their business function
* Comply with all industry security standards relevant to your business such as PCI DSS, HITRUST, ISO27001, and SSAE 18 --- LOL to this whole line item.
* Maintain an up-to-date diagram indicating how sensitive data reaches your systems and where it ends up being stored
There is then a longer list of controls that I think most practitioners would say are good things, but that aren't always P1-prioritized (for good reason, to make way for more important things). CSP headers, SLSA level 1 builds, media sanitization policies; these things are situationally important.
I think an opinionated checklist is a fine thing, but when you call something the "minimal viable" standard, you set yourself up to explain how lots of well-run companies are viable without these things.
Or closer to recommended practices. As these are definitely not minimal since there are many products that won’t meet them yet are still viable.
It’s also funny how cyber consultants seem more apt to add “required” thing that really tend to increase their bill hours. Like how accounting firms love Sarbanes Oxley and like to add in new rules that they claim are necessary.
Edit: court ruling (shortened SOX) has to do with security compliance for public companies. After Enron and other scams
I mean the list isn't for savvy security teams. Or, I'd say, not even for companies with any kind of security team.
For the typical startup with zero people with any security experience, it's something they can do. If they get to a point of hiring a security team, they're past the point of needing such a starter list.
> Comply with all industry security standards relevant to your business such as PCI DSS, HITRUST, ISO27001, and SSAE 18 --- LOL to this whole line item.
That's silly to say. You must comply with regulations that apply to your industry. Try setting up a bank and just LOL at banking regulations (SBF?). Of course, if none apply to your industry, ignore them all.
Yeah this should just be the list. But then it wouldn’t be it’s own thing. It would just being SOC2 or NIST-800 certified.
I stopped reading at this point, completely disillusion for anything _minimally viable_ it's 2023 and I've still seen large companies with credentials and passwords committed into their code...
I’ve got some experience as Head of Cyber/InfoSec at a couple of startups/scaleups and I see this as potentially useful in a bunch of ways if broadly adopted. It establishes a baseline both for our own products and our suppliers, attests to stuff that actually affects how we integrate with and operate a third party product (ISO 27001 gives me no clue whether you support SSO, have viable logs that I can actually ship to the SIEM etc.) and hopefully simplifies both the due diligence we do on our suppliers as well as that done on us by our customers by allowing us to reduce a good chunk of the product feature questions to “does your product meet the MVSP standard”.
There was a pretty useful discussion of it on the Google Cloud Security podcast a while ago: https://cloud.withgoogle.com/cloudsecurity/podcast/ep114-min...
My criticism of this is that you can secure the product with solid defense in depth on the enterprise/infra side in a way that is:
- minimally invasive to devs
- aligns with what is likely there to slot into - what SAST tool can you pay for that early on at a startup, let alone slot into a non-existent DevOps pipeline. $1-200k SIEMs existing at a series A/B that prob has 1-2 sec engs?
- get a WAF for a bandaid fix for the appsec problems initially until you can hire product security folks
But when you look at some key aspects of that infra/biz app security journey at a largely cloud based young business - like taking more control of IAM across SaaS, bringing together activity from those multiple cloud systems and creating alerts that actually matter - that all involves knowing your vendor supports SSO, possibly also SCIM, has logs that contain relevant data that can be exported/streamed into a SIEM. Some of the stuff this standard demands is going to be fairly core to securing business infra. I just get bored talking to vendors that want to wave an ISO cert at me as if that somehow makes it ok they don’t give me access to audit logs.
Ultimately it boils down to:
> Convince whoever is asking you are following whatever process they are asking about so they can check a box.
They call it security theatre for a reason. Of course if you don't want to be lying, well, you gotta implement a lot of this stuff as at least written policies. Also, auditors have a tendency to ask for evidence...
Google (one of the main sponsors) and other cloud-hosting vendors can essentially say, "Sure, you can cobble all of this together yourself -- or you can buy our services with much of this already baked in."
This list looks like it was written by people who are not paying for anything, so they don't understand the tradeoffs, costs and compromises. I can assure you things look quite different when you are a solo founder running the business and you have to pay for everything. Technical stuff doesn't change, so recommendations like "Do not limit the permitted characters that can be used in passwords" still hold, but "Contract a security vendor to perform annual, comprehensive penetration tests on your systems"?
Also, "Publish the point of contact for security reports on your website" and "Respond to security reports within a reasonable time frame" — I'm already spending too much time responding to "security researchers" trying to shake me by sending (rather silly and generic) "vulnerability reports". Some of them follow up by asking if they can "publish this on their social media channels" if I don't respond. I really don't need more "security reports" for my website.
That's the important word there (emphasis mine). I'm doing an MVP now for a client for a custom app (used internally by that client, i.e. only employers are users), and my recommendation for security is "I'll use a simple XOR symmetric stream cipher for now. We can switch to TLS[1] if the system turns out to be useful".
Yeah, cut corners on other things as well - they just want to see how viable this thing is in practice.
[1] While TLS is nice, certificate management becomes a pain in the rear for remote devices.
I don't get it. Why do we need all these additions? Perhaps it is me who does not understand the meaning of "Viable". MVP is not a restricted, well-defined end-goal. "Viable" means viable for your case. You can fill in whatever you want. It does not mean "Viable" without security or "Viable" without "Marketability".
But zero security is exactly what most startups see as viable.
I remember one PM at a startup saying "Well hire a marketing firm to create flowery language for our website asserting how ultra military grade secure we are. And that's it. We're not going to spend a penny implementing any of that security geekery you people talk about."
That doesn't let them off the hook, but all that process overhead can be a killer.
I worked for a company that was PCI DSS, and it often made it impossible to get any work done (to be fair, it had more to do with how they implemented it, than the standard, itself).
But I agree that security needs to be Job One for everyone, regardless of size.
My minimum for a public-facing MVP:
- All services use HTTPS to talk to users and each other.
- High standard of password encryption (or prefer to use something like Cognito or Firebase).
- Plan for GDPR compliance (not exactly security related, but in the wheelhouse. The GDPR grifters come out of the woodwork quickly, so you need all the popups and account deletion stuff from day 1 if you are releasing to the EU).
- QA specifically for security: users can't access each other files, authentication controls work, etc.
- Don't store or handle credit cards - use a vendor like Stripe.
- Ensure all dev tools enforce 2FA where possible (GitHub, AWS, etc.).
- A basic backup system.
Then post-MVP, start working on the following:
- centralised logging.
- dependency patching plan.
- etc
Alternatively, if your customers will make one-off or infrequent payments, you might want to consider accepting cheques, which can be made electronically through a system like BACS (note that every country has different systems available). Cheques (checks) are a 'customer pushes' method as opposed to cards, which are a 'seller pulls' method. Accepting international payments makes transfers somewhat more difficult, and cheques always assume a higher level of competence on the part of the customer than cards do; however, neither are likely to be a problem for B2B products.
> Ensure all dev tools enforce 2FA where possible (GitHub, AWS, etc.).
Two things that are critical to remember are that 1: most forms of 2FA don't improve security and 2: many services will refuse to restore accounts based on trust or some other form of evidence if 2FA was enabled, where they would otherwise.
Challenge-based forms of 2FA such as TOTP or FIDO are better than SMS, as latter can be intercepted in transit. Calculating the response to a challenge on a separate physical device to the one being authenticated is additional benefit that FIDO2 usually has. Side-note: if you need to authenticate customers, allow them to choose at least TOTP; try not to even provide SMS 2FA.
As for the lack of account recovery, this can be a benefit, but only if you have robust procedures to make sure employees' credentials (like TOTP codes) are copied to sites accessible by other employees; this effectively means things like buying fireproof safes if you are doing it properly. Revocation of credentials by other employees is just as important as recovery. All this is to say that 2FA is not something that you can just toggle a switch for to make your company secure; it is worthy of a company-wide strategy.
Disclaimer: I used to work with the founder, he’s great.
Sure, although minimum viable and enterprise-ready seems like an oxymoron to me.
Step one: define MVP.
Step two: add minimum enterprise security, minimum enterprise scalability, minimum enterprise legal compliance, minimum enterprise cost controls, minimum social responsibility ...
Step three: why the hell is MVP 28 months late?
But seriously, I think the problem is in the name. This is not a set of best practices for any kind of startup but rather a public checklist that anyone looking forward to provide some kind of service or product to any of the contributors (i.e. Google, Salesforce, Okta, or even other companies subscribing to this extreme SecOps cargo cult) must comply.
* Check you're storing secrets safely and not in the code
* Run a automated security scanner to check for the OWASP top 10
* Confirm your endpoints have correct auth/permission checks, and that all debug flags are off
* Ensure databases are protected (eg. Not exposed to the internet or heavily restricted)
* Enable 2FA for everyone in the company and use a password manager
* Check what are the data protection laws and disclosure sla of your country
An old product attempt of mine was a threat model wizard that generated a simple deck with some very clear viz of the model and threat exposure addressed to all concerned parties - and then backlog issues and themes and epics for implementing or verifying the controls it implied. It reduced the checklist/risk assessment process to about an hour from the weeks of spreadsheet work that was a lot of peoples jobs, and it put security into the product/project dev process. Pretty much aha.io (the product management tool) but for security.
What I learned from it is a) self-assessments weren't valuable to large org customers (who need to be able to say they were told by a 3rd party, so it's not like product management that way) b) the need of small orgs / startups was a standard compliance checklist to make standard assertions to potential customers as part of vendor onboarding, and a faster/better model didn't get them that standard. c) most security people trade on their insights, and codifying their ontology even just to speed it up undermined their leverage in their orgs.
This checklist or something like it might actually meet the real needs of some startups who must make templated assertions to customers for vendor onboarding, but most startups would be lucky and probably pretty chuffed if they actually had something someone wanted to steal.
I would love to hear what others here consider important to make a minimally viable and secure product in a startup from the starting as a startup.
How much of this list is hard needs, soft needs, beneficial, and nice to have preferences/interpretations?
Many startups wouldn't make it to the start line fulfilling this list. If I could pick a solution architect's brain looking at this, I'd be curious to know what ways there are to satisfy these through architecture, design considerations, or using particular parts of particular platforms?
English is not my first language and I’m not a security expert, but this description seems a bit misleading.
The “Origin header” part should be left out. You don’t check where it’s comming from (you can’t know). You either send a unique token back and forth, or you defer the issue to the browser (cookies with Same Site strict/lax).
And it’s not “requests”, it’s “unsafe requests” AKA mutations (POST, PUT, PATCH, DELETE). It’s not just a nitpick. If your GETs are not safe, you might cause deeper issues.
Some previous discussion: https://news.ycombinator.com/item?id=29100400
