We built the vuln scanner we wanted to use but couldn't find.
Introducing Vuln Scans in Detail!
It uses the same abliterated models attackers are using to create havoc across the internet right now.
Under the hood we are:
- Spinning up your dev environment in a sandbox
- Throwing an enormous amount of abliterated model tokens at trying to hack your code
- Cross-checking your supply chain with cve/.now from Sazabi, to find any supply chain vulns that are exploitable in your app
This is all happening within a cloud dev environment. We will not hack your prod system.
The abliterated models are scary good. We're finding nasty issues in every repo we scan. We've found 13k vulns so far, and the majority are already fixed.
We see the current moment as an all-hands-on-deck situation. Agent swarms are stringing together minor-seeming vulns across different apps to hack systems in new and unexpected ways. We all need to find and patch an enormous number of vulns, fast, or the internet is in trouble.
You don't even have to be a target to get hacked right now. Your entire product might be a stepping stone in an attack aimed somewhere else.
We can remediate findings for you (i.e. send you PRs that fix the vulns), or you can consume them via your agent of choice, or in your ticket tracker of choice.
If you need help or advice on getting your system secure fast, please get in touch. We have helped teams of all sizes adapt to the current moment. DM me.
We're delivering these scans at-cost, because it's all-hands-on-deck. We hope you'll get one today.
We spent two days trying to hack ourselves with abliterated models. The biggest risks we found, by far, were side channels into our infra, outside the application code.
Stuff like:
- The oneoff dev env we set up to test out Coder that had way too much access
- CI runners had tailnet access for no good reason
We’ve scanned our application code many times over, but there were a lot of scary risks hiding in these side doors and third party vectors. All problems that were very easy to mitigate once we were aware of them.
Many of the high-profile hacks from the past two weeks happened this way. Application code held up fine, but the company had a massive breach due to a vulnerability in a vibecoded internal tool or an unpinned CI dependency or third party js or …
You’ll see us incorporating this pov into our vuln scans. (And of course we’ll continue perpetually trying to hack ourselves!)
Also: we tried to hack ourselves with regular non-abliterated deepseek too. We got a lot of model refusals and a worse result.
I’ve seen some people asking whether the abliterated models are actually that much more amoral in a real-world setting. The answer is definitely yes.
We spent two days trying to hack ourselves with abliterated models. The biggest risks we found, by far, were side channels into our infra, outside the application code.
Stuff like:
- The oneoff dev env we set up to test out Coder that had way too much access
- CI runners had tailnet access for no good reason
We’ve scanned our application code many times over, but there were a lot of scary risks hiding in these side doors and third party vectors. All problems that were very easy to mitigate once we were aware of them.
Many of the high-profile hacks from the past two weeks happened this way. Application code held up fine, but the company had a massive breach due to a vulnerability in a vibecoded internal tool or an unpinned CI dependency or third party js or …
You’ll see us incorporating this pov into our vuln scans. (And of course we’ll continue perpetually trying to hack ourselves!)
@nateberkopec White box, which was key for discovering these high risk vectors that didn't happen to be exploitable but which we were very glad to know about and preemptively remove.
I can control when an npm dep changes. You especially do not want mechanisms that let third parties change the code that runs in your product without you even approving the change.
If someone hacks Sentry and publishes a bad SDK version, I can protect myself a dozen ways, including via something like Socket. When Optimizely got hacked yesterday, a lot of people were just going to be hosed, because Optimizely doesn't even permit SRI.
I would strongly consider uninstalling any third party js right now if I were you. Especially if it comes from a startup.
Analytics, monitoring, etc are all massive site takeover risks. If you use a tracking script from a third party SaaS, then if they get hacked the hackers have total control over your site too.
This is what took down Polymarket today. They used a tiny script from a SaaS that tracks page load speed. That company got hacked, and the hackers subsequently stole millions from Polymarket users.
If you want to use analytics safely, I would use a product that lets you pin the tracking script and control when it changes. Most of the mature products support this, e.g. Sentry, Amplitude, PostHog, Mixpanel, Heap. If not, you need to trust the company behind it a lot...
Railway got hacked last week. Railway! That's a very competent team, with particularly strong infra chops. The agent swarms are on another level.
A typical analytics company is nowhere near as sophisticated. You should ask yourself: do I trust this company's security more than Railway? I was an analytics CTO in a past life, and I will be the first to admit: we had some great people but we were not on this level.
We (Detail) just uninstalled our session replay, which was a product I loved but is not worth the risk right now.
I hate to write this because I spent a good chunk of my 20s benefitting from product teams taking a chance on a nothing startup and installing our js on their site. But in the past two weeks this risk vector has exploded. It's better for the ecosystem if we move away from third party js as fast as we can.
Optimizely got hacked, which means any website that uses Optimizely could get taken over. Hundreds so far.
Once attackers had access to every Optimizely customer's website, they were able to:
- skim credit cards by recording user input
- hijack logins by stealing cookies
- steal crypto by redirecting transfers to their own accounts
The best time to remove third party js from your website was a month ago, the second best time is today. Or use subresource integrity to control when any changes happen. Analytics, monitoring, A/B testing, it all needs to be locked down.
If we want to get through the next six months, we should be as aggressive as possible about avoiding these kinds of amplification vectors that allow a hacker to breach one moderately defended system and use it to hack thousands of other products.
I would strongly consider uninstalling any third party js right now if I were you. Especially if it comes from a startup.
Analytics, monitoring, etc are all massive site takeover risks. If you use a tracking script from a third party SaaS, then if they get hacked the hackers have total control over your site too.
This is what took down Polymarket today. They used a tiny script from a SaaS that tracks page load speed. That company got hacked, and the hackers subsequently stole millions from Polymarket users.
If you want to use analytics safely, I would use a product that lets you pin the tracking script and control when it changes. Most of the mature products support this, e.g. Sentry, Amplitude, PostHog, Mixpanel, Heap. If not, you need to trust the company behind it a lot...
Railway got hacked last week. Railway! That's a very competent team, with particularly strong infra chops. The agent swarms are on another level.
A typical analytics company is nowhere near as sophisticated. You should ask yourself: do I trust this company's security more than Railway? I was an analytics CTO in a past life, and I will be the first to admit: we had some great people but we were not on this level.
We (Detail) just uninstalled our session replay, which was a product I loved but is not worth the risk right now.
I hate to write this because I spent a good chunk of my 20s benefitting from product teams taking a chance on a nothing startup and installing our js on their site. But in the past two weeks this risk vector has exploded. It's better for the ecosystem if we move away from third party js as fast as we can.
@aaronjmars Our cost per medium is about $17.50 right now.
We have a lot of ways to bring this down, but this is already at the point where you really don't want $500 worth of medium+ vulns lying around in your product, and $500 is a small price to pay to fix them before hackers find them.
One week since launching Vuln Scans in Detail, here's what we're seeing:
- One exploitable vuln found per $3.45 of customer spend
- Average of 134 exploitable vulns (!) per repo
- Average of 7 exploitable supply chain CVEs per repo
- One high/crit vuln per $200 of spend
- Of the high/crit vulns, the strong majority have been broken auth, e.g. users can access info from other accounts, or even non-users can access user info
- Also a lot of injections (e.g. an unsanitized field lets hackers run shell commands), sensitive data exposure (e.g. logging secrets in error cases), and request forgery
Tl;dr: it is much, much cheaper to find vulns when you have access to the code. We need to take advantage of this fact if we want to keep software from getting hacked.
@0xdoppo It really does. We found this a while ago when investigating ways we could characterize the bugs that come out of agents. https://t.co/0eHxolg2La
One week since launching Vuln Scans in Detail, here's what we're seeing:
- One exploitable vuln found per $3.45 of customer spend
- Average of 134 exploitable vulns (!) per repo
- Average of 7 exploitable supply chain CVEs per repo
- One high/crit vuln per $200 of spend
- Of the high/crit vulns, the strong majority have been broken auth, e.g. users can access info from other accounts, or even non-users can access user info
- Also a lot of injections (e.g. an unsanitized field lets hackers run shell commands), sensitive data exposure (e.g. logging secrets in error cases), and request forgery
Tl;dr: it is much, much cheaper to find vulns when you have access to the code. We need to take advantage of this fact if we want to keep software from getting hacked.
@AIQuanting Regression risk for these vulns tends to be very low – a lot of the fixes are very local – so yes many people are accepting that tradeoff as the most practical way to get through the next few months without a major hack.