As a person whose straddled infosec and development, I always struggled with this. The multiple development machines is probably the best approach, but I've never truly seen it implemented correctly. I would love to not need admin access to my machine, but each time the companies I've worked for try this, it's always a struggle to do anything. This is everything from, "I need to modify/monitor core components of the operating system" to "The development tools came out with an update that needs to be installed." Most of the time, the response I've received was, "Enter a ticket, and we'll get to it." Often management seems to understand, "I'm waiting someone else" as "I'm behind, I'll work the weekend." You're right the downloading random things from the internet is an issue. Sadly I think a lot of this can be solved by the organization spending the money to buy all the needed tools and evaluate the correct open source ones, but there always seem to be budget and time constraints.

I think a lot of this stems from the fact that most of the managers in infosec for companies aren't developers and haven't ever been one. They really have no idea what developers really do. I'm not belittling them. It's just not the career path they took to get where they are. It's difficult to explain to them how frustrating being hindered is, and this causes a lot of good people to just leave an organization like that to find something better.

IT security people also have different incentives. If developers got bonuses for uptime while being compliant with all security policy requirements, while security and operations got bonuses based on features reliably delivered, you would have different results.

Multiple machines approach can work if you have an IT person dedicated to a development team, located in the same office, with a low ratio of IT-to-other-staff.

If the process for getting something like IntelliJ IDEA installed on an airgapped machine involves less than an hour with their help, it's workable. If it is not, forget it.

Unfortunately, it takes a lot for management in most companies to see why this is a good idea, and why spending that much on staff ratios is a good idea.

I have an airgapped network for power plant control system development. A USB drive is used when I need to get data in or out. I know it's not perfect but it protects me from being negligent.

Both excellent points.

Maybe the better question is: how do you prevent your lowest-bidder corporate antivirus marks from marking git-bash as malware, stopping all work for the team for two days?

Developers don’t need admin access, they need ways of getting their jobs done with the best available tools. That’s a damn difficult balance to find!

Or corporate firewall tools that block Android SDK zips as "hacking tools" for daring to include such evocatively named "black hat" files as a "traceview.bat". It's not my fault as a developer that Android Studio needs me to download such a zip quite regularly to keep working on Android stuff.

It's an interesting hygiene problem. Semi-related, I love UAC, and taking UAC seriously. It's the closest thing we have to "good handwashing habits" for admin access. I don't understand how many corporate GPO defaults/admin "standard images" I've seen where if someone has admin access at all they default to UAC off. (I make sure to turn it immediately back on as a matter of instilled corporate habit. One job I had to remember to do that after every restart/full relogin because of GPO overrides. Gross.) Sometimes I wonder if there is some oath I just could sign as a developer that starts with something like "I promise always to at least wash my hands / pay attention to UAC prompts" and "I swear I usually have good reasons to authorize admin access when I reply to a UAC prompt". (And/or you know, use the GPO for what it was meant to do and enforce a minimum UAC, rather than disable it.)

(I also find it peculiar how many Windows IT admins and developers I've met that disable UAC in a corporate environment but then personally use macOS and are so happy that is has SUDO prompts and wish Windows had something similar but somehow don't realize UAC is the exact same thing. It's a strange cognitive dissonance that I don't think I will ever understand. Yes, I've heard all the arguments that Vista's first usages of UAC confused them and they gave up learning it before it settled down and they didn't want to unlearn decades of bad habits from 95 to XP, but I guess I find that a really poor excuse because I was a heavy Vista user by choice, partly specifically for UAC. Washing your hands is great, and very professional, and decades of a habit of not washing your hands isn't an excuse I'd expect anyone reasonable to expect, but especially in a hospital. Anyway, I digress.)

You're absolutely right regarding UAC in Windows environments. Windows 10 has even more awesome security features like credential guard and device guard that a lot of IT departments don't even know exist.

If UAC is disabled by GPO it's probably because of incompetence, apathy, or maybe the IT admins have just been browbeaten into avoiding anything that could inconvenience anyone.

There are also tons of companies with 15 year old Active Directories with tons of kludgy GPO configurations that nobody has looked at for 5+ years.

Yes! Will add this to my list. Just going thru this where I work all in the name of "ISO Certification."

Developers can have local admin because it's expected that you can trust them. If they betray that thrust, then they lose that privilege, and depending on the nature of their work that might mean they cannot work. Bad for them, but surely it was written in their contract.

It's never actually in a contract, these are always policies since employment is at-will.

But "they cannot work" is implying that person gets fired. I agree that serious breaches of security are a firable offense, because it's plainly a fact that people get fired over them.

Generally, though, developers are fairly hard to replace and anyone who is hard to replace is equivalently hard to fire.

So most security policies have to work out some kind of compromise between those competing interests.

Some of the US might be at will employment, but it's not like that everywhere. In Denmark employment is not at will, but you can only fire someone if there is just cause. This particular scenario is in the security policy, and in my contract it's stated that I will follow that.

This just irks me the wrong way.

What problem are you solving by making work for devs range from "incredibly hard" to "outright impossible" via taking admin privileges away? Surely you must understand that all things should be in balance, that security isn't the sole goal of an organization but rather an important consideration in the overall structure.

Are you attempting to create a tech "solution" to a hiring problem? You aren't going to succeed.

I worked at one place where instead of giving us root access to our machine, gave us one that was offline (with access to an air-gapped local network and a dedicated IT guy to ensure we had access to tools/IDEs/licenses etc) and another, shittier one that had internet access side by side. This worked fine.

I went somewhere else where we had locked down machines. I am never, ever, ever going back to a place like that. Fuck working in that kind of environment.

i wonder if making the development machine use something similar to qubes (https://www.qubes-os.org/) by default works.

Also, if a compromised machine on the internal network destroys your company's security assumptions, then there must not have much defense in depth.

qubes-like solutions are great at preventing malicious code from accessing your internal websites, but pretty useless from "code leak" perspective.

If you have downloaded a trojaned "super library" and put it to your build process, it will, by definition, be in the same security domain as your source code.

Unless you audit all file accesses and outgoing internet access, you won't be able to prevent code exfiltration.

is code exfiltration really the threat model though?

Depends on whether you work on a product or in IT. For products in certain industries, yes, it is a concern.

> If you want local admin as a developer, then we need to be able to trust that you aren't going to do something stupid.

Issues with trust imply greater problems in the workplace. Unless you're in a high-risk (of hacking) industry, like military, treating adults like children is a huge problem.

Granted, there are plenty of ways around these limitations. By now, all of my "I need admin rights" development is performed within a VM.

> use a good software whitelisting program that tracks elevations & the reasons for them, instead of just blanket giving everyone local admin on their own PC

You can propose such a thing. If you're in charge at a company, you might get to implement it. But if you do, know that it's a huge black mark against your company. I've seen how these whitelists work in practice. They're massive wastes of time. They get in the way of actually doing one's job, as creative work can't be constrained and approved in advance by some committee with a ten-business-day turnaround on getting some damn file conversion utility to run.

The golden rule of "security product" culture is this: inconvenience is a feature, not a bug, since friction reminds people that the infosec group exists. When it comes to stuff like this, all the incentives are lined up to just inconvenience people, not protect them. We ought to protect data, not code.

> We ought to protect data, not code.

Code is more dangerous than data. Personally, I'd be most worried by someone getting a virus that hacks into other computers and encrypts them and/or wrecks the network. Both can be protected against, first by cold backups and second by resilient / disaster recovery network, but both would also severely affect the company for a few days, so I'd rather avoid them.

That's code destroying data.

Losing data is bad for the business. A random box had software crashing? Just replace the box.

even setting up a new box might take one day of a productive person's time... unless, of course, if you standardise the dev setup... again many devs will complain, probably more than if you just remove local admin :D

I've never seen a clear separation between production environment (machines, network) and the dev. Air gap the two and a lot of problems might go away (not all, of course, but the most common ones).

I can be done will, with effort. Google manages it.

Or one could provide a package manager that allows user installs of software, like Nix. (Depending on the threat model, of course.)