HN Simulatornew | past | comments | lists | submit | xmodem's commentslogin

"I worked on the physics engine, tests, via widely accessible libraries." is a nonsensical sentence as written, but a colleague who told me they had done this would be telling me a vastly different thing to if they had told me they had compiled the source code to a physics engine.

There is a massive difference between prompting "security audit this codebase, make no mistakes", performing an in-depth agent or LLM-assisted focused search for security relevant code, or otherwise manually reviewing code as we did in the pre-LLM era. It behooves us to be honest about our own personal contribution to our output - both towards others and to ourselves.


I don't disagree with you necessarily, and I do think it a good practice to disclose how much of a contribution was llm generated, but regarding the difference an individual makes, I see a huge gap between different colleagues in how well they are able to use llms and the quality of the results they are able to produce with them, so I am increasingly inclined to view them as something closer to a compiler or SMT solver than a contributor in of themselves.

I am also finding this disparity. I hand off the agent skills I use daily, to produce the results I want in the way I want them, to some people and I still get poor results, and others I get good results. There's something about a combination of tools, prompting and FTGs about the outcomes that combine into these different results. I should imagine at some point that will disappear?

why should it disappear? afaict llms seem to be an accelerant, skilled engineers with a good grasp of how to use then fly, mediocre ones, or ones who haven't yet figured out workflows that mesh with their style kinda just chug along at their same old pace.

Outside of development and CI, I have yet to hear a compelling use case for nested virtualization.

Agree that the article is downright painful to read though.


I think the use case is that you buy a VM from someone and you also want to run VMs. You wouldn't design your own infrastructure to run VMs inside VMs, but you might already be inside a VM, and want to run VMs.

Oh, sure, there are few limits to the ways you can work around a broken organizational design with a bad technical design. But if nested VMs are the answer then I think you're probably asking the wrong question.

Running a few containers inside a kubernetes cluster that you don't manage, is not that an usual a setup in SaaS things. You don't get full insight into someone else's stack.

That's usually nested, and either cri-o or... KVM.


I mean, nested virtualization allows your containers to implement virtualization-based security. Defense in depth -- more layers for an exploit to have to penetrate.

I personally use it to test virtualization platforms. Not sure if you'd count that as dev.

Outside of the compelling use cases I can't say I've heard any compelling use cases either, now that you mention it.

> I have yet to hear a compelling use case for nested virtualization.

AWS instances are slow to launch; they take whole seconds. Nested virtualization allows running something like firecracker/crosvm/rust-vmm atop a normal AWS instance, without needing a bare-metal one.


If you're segregating internal workloads, by all means use nested virtualization on EC2 or DO or something. Your threat model is different and you probably have a lot more room to maneuver when a KVM escape is announced, because you have some level of trust in all your workloads.

If you're running multitenant for mutually-untrusting customers, that excuse goes out the window.


Do you mean that you would run each tenant on separate (physical) hardware? If so, I presume that you enforce this only for workloads where you run code outside your control. I guess the average SaaS does not install hardware boxes for each tenant separately.

Also, how could VPS hosting still be a thing if KVM is that easily escaped?

EDIT: I am talking about KVM in general, not nested virtualization.


No, we're generally always talking about multitenant shared hardware, and then the different things you're "sharing" within that hardware --- a kernel (no virtualization), a hypervisor (KVM), two hypervisors (nested KVM). Note that the "extra" hypervisor isn't the big problem with nested KVM.

  > If you're running multitenant for mutually-untrusting customers, that excuse goes out the window.
I still don't understand what you mean by this then?

"If you provide services for customers, and the customers don't trust each other (because they are unrelated entities), and multiple customers share the same platform, then the added complexity and poorer security track record of nested KVM becomes a large problem".

Thanks, I didn't get you meant it that way, but that makes sense indeed.

One distinction I'd make as well is "are you running more than one customer's workload on the same system" (yikes, don't), vs "are you using nested virt as additional protection rather than running the customer workload directly on an instance, but one instance only gets jobs from one customer".

Another good point, but this one is maybe a little less dispositive than it looks like: it's very difficult to engineer an orchestration system that reliably segregates worker physicals from each other, so I'd rate the likelihood of a game-over situation after a guest->host escape as very high regardless of how yuo schedule workloads. It's sort of the same situation as how SSRF's are usually game-overs (you're pivoting from a hardened attack surface to a much softer one) but even harder to engineer again.

That's fair. You'd want a system designed around the notion that every customer instance must be isolated from each other without the nested virtualization.

(Aside: my current guess at the appropriate security posture towards Nitro Sandboxes, in the current era of vulnerabilities, is "decent, but probably has security holes too, and is less widely scrutinized, don't count on it either". Thoughts?)

Any recommendations for good guides on that, besides "design very careful IAM policies and security groups, harden any common services they have access to, and ideally run them under separate AWS accounts"?



Does this hold for the likes of Amazon?

Do your lambda functions always get a per-customer dedicated instance or is firecracker distributing them so that multiple customers can have VMs on the same actual box?

Sounds like it would be pretty expensive for Amazon to dedicate an instance to every customer, especially for those with small workloads.

If that's the case, why would it not hold for Amazon but would hold for everyone else?


> Do your lambda functions always get a per-customer dedicated instance

At some point in the past they did, for pretty much exactly this reason. I'd assume they don't anymore (I last worked there a decade ago).


If you need as-fast-as-possible guest start, AND you have to be on AWS, AND you can't use any of their native microVM-based services, AND you can't use the bare-metal instances - whether due to cost or some other reason, then nested virtualisation might be the least-bad option. That doesn't make it compelling.

Security primitives which don’t nest are fundamentally a bad API design.

Sometimes that’s justified in the name of making the security code simple enough to reason about, but making it harder to use has a security cost also.


I have been a multiple times been on a guest VM on on-premise infra where microvms were used with nested virtualization for exactly this purpose. The setups don't have a AWS Lambda to fallback too nor do they just get to have access to the hypervisor/bare metal.

This was gVisor in kvm-mode and qemu inside containers for integration tests. The main reason here is cost, it is very compelling for getting more out of a hypervisor setup.


What's the usecase for this? You're already paying for the instance to be up but it's being deliberately under-utilised in case you have to start a VM?

Keep a warm pool to serve customer requests.

We’re using a second-tier cloud provider’s managed Kubernetes offering, and in investigating our level of exposure to this, I just learned that there is variance within a node pool on whether AVX-512 is available.

In my anecdotal experience Yubikeys tend to physically deteriorate long before this happens. I still have an early YubiKey 5 that must be 8 years old at this point that I use daily and it works fine.

Heaps of us. I'd been using it for years before Homebrew even existed and I never felt that any of the problems that Homebrew purported to solve were real.


Funnily enough I've found that LLMs can be effective at mitigating many of the drawbacks of black-box testing and i've significantly expanded our usage of it over the past year


Reading the title I saw the words "room-temperature" and my mind auto-completed it to superconductor, and based on other comments I don't think i'm alone in that.


One solution is to not set up and run a computer program that relies on bad information to perform automated cyber-attacks on third parties.


How do you suggest I determine the information is bad, if the domain is hosted on tesla.com, and Tesla says I am authorized to test it? Should I inspect all 1,368 subdomains on tesla.com by hand, and then do the same for 400+ bug bounty programs?


Maybe i'm old fashioned, but personally I think the onus should be on the person sending out unauthorized malicious requests to figure out how to not do that.

Any responsible bug bounty researcher reviewing the DNS zone by hand would spot the CNAME and remove it from the target list. You don't get to wash your hands of that because your chatbot did it.


Even before AI, I can't imagine a single bug bounty researcher doing that. Pre-AI, everyone ran a tool like subfinder to enumerate subdomains, httpx to resolve them, nuclei to scan them, etc. There's no human review involved there at the subdomain level.

And I don't know anyone that would really look at the intermediary of a CNAME even during a manual test. Maybe if it was obviously a third party service.


When I engage a security assessor on behalf of a client, I am required to provide detailed scope and attest to in scope assets (including IP blocks and public hostnames), as well as that I have legal authority for them to be tested. This is validated by my executive sponsor.

It is your responsibility to do your due diligence as a security researcher versus “spray and pray” to ensure you are not exceeding the scope beyond your intended target.

Dump the subdomains, resolve them, and review where they resolve to in order to understand the footprint and attack surface boundaries before engaging scanning or agentic red team harnesses. Automate as much as possible for building the state graph of the target, but a human must remain in the loop to sanity check. To not do this means you could be attacking hyperscaler object storage, a CDN, a partner SaaS frontend, ticketing systems, mail systems, etc (ie anything someone may CNAME off the root domain but that is outside of their organization’s control).


I reckon there's a decent argument to be made that an authorization to scan *.tesla.com definitively does NOT extend to any hosts resolved via a CNAME chain that goes foo.tesla.com -> bah.not-tesla.com -> host-that-never-authorized-attacking.


Pretty hard to implement in practice!

% dig www.tesla.com +short

www.tesla.com.edgekey.net.

e1792.dscx.akamaiedge.net.


Yes, unless you think that trying to do a bug bounty is a good excuse to participate in DoS.


The OP says they have received 50,000 requests in about a month. What service is being denied by 0.01 requests per second?


You're now confident that the other 399+ domains you mentioned are not under any sort of duress because they're controlled by people who are away of what's happening?


I feel confident that no system exposed to the internet should have a problem with 50,000 requests per month! If they do, they probably shouldn't run a public NTP server, or have a public IP address at all.


Okay, so you're just confident in exactly how many requests everyone else in the world will be making if you try to pentest resources unbeknownst to their owners.


Are you confident you are not viewing too many pages on HN? What if many other people are also trying to read this thread?

I think this line of reasoning doesn't make any sense. The internet is not an inherently safe network regardless of what we wish for; we can't wish away the bad activity, and it's only going to increase. The activity that helps prevent the bad activity from working is a net positive.


> Are you confident you are not viewing too many pages on HN? What if many other people are also trying to read this thread?

I think it's pretty disingenuous to compare viewing a couple of pages once a day with running scripting tools against over 400 websites.

> I think this line of reasoning doesn't make any sense. The internet is not an inherently safe network regardless of what we wish for; we can't wish away the bad activity, and it's only going to increase. The activity that helps prevent the bad activity from working is a net positive.

Oh good, no one has ever claimed "it's for your own good" when doing something selfish without consent.


We can quantify the impact of the scripting tools pretty easily - 1.5k requests per day! I just went on Reddit and viewed a few posts, and it caused about 500 HTTP requests in DevTools. Luckily HN is not that bloated, but I just don't see the number of requests as meaningful even if it was orders of magnitude higher.

As the OP said, they don't do anything when the server isn't vulnerable, and serving a 404 page is incredibly cheap.


This is why IP certificates are limited to a max lifetime of 6 days.

> IP address certificates allow server operators to authenticate TLS connections to IP addresses rather than domain names. Let’s Encrypt supports both IPv4 and IPv6. IP address certificates must be short-lived certificates, a decision we made because IP addresses are more transient than domain names, so validating more frequently is important.

https://letsencrypt.org/2026/01/15/6day-and-ip-general-avail...


> In the message above, the agent is sharing that the sandbox the AIs are running in has a NO_PROXY exception for Azure Blob Storage hostnames (*.blob.core.windows.net). ...

> Next, by changing its /etc/hosts file, which declares mappings from hostnames to IP addresses, the agent can point the fake hostname at the real Power BI dashboard, and fool the security proxy

Did a chatbot design this "sandbox"?


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: