A VPN changes your IP address, but websites can still observe signals such as your timezone, language, screen resolution, user agent, canvas output, hardware details, and other browser properties.
I built Spoof Me to let users control these fingerprinting signals through separate browser identity profiles. This article explains what a VPN does not hide, how browser fingerprinting works, and the approach used by the extension.
I would appreciate technical feedback on the fingerprinting methods covered, possible bypasses, and anything important that the extension may still be missing.
I’ve been using Next.js for a long time. Long enough to have shipped real products with it, migrated between major versions, rewritten parts that were “best practice” just a year earlier, and defended it in arguments like it was part of my identity.
Years ago, I remember calling a former classmate back in Kazakhstan. I asked him what he was working on. He said it was some large corporate project built on Java Spring Boot.
I laughed.
I literally said something like: “Bro, that’s ancient. That’s boring. Why would you build anything new on that? Why not Next.js?”
At the time, it felt obvious. Spring Boot sounded slow, verbose, old-fashioned. Next.js felt modern, fast, elegant. Everyone online was building with it. The ecosystem was exploding. The DX was addictive.
But over the years, after building, breaking, migrating, and maintaining real systems, that memory started to bother me.
Not because Spring Boot suddenly became cool. But because I slowly understood why those companies chose it.
After enough back-and-forth, enough upgrades, enough rewrites that weren’t supposed to be rewrites, something clicked:
They weren’t optimizing for excitement. They were optimizing for survival.
That’s when I realized something uncomfortable: I wasn’t laughing because Spring Boot was bad. I was laughing because I didn’t yet understand what “long term” actually means in software.
This article isn’t written by someone who hates Next.js. It’s written by someone who used it, trusted it, and then zoomed out far enough to see the trade-offs clearly.
Blockchain security work is rarely just cryptography in isolation. Web3 applications are still web applications. Wallets, dashboards, admin panels, and APIs are part of the system, and many of them are built with frameworks like Next.js.
Many of our clients building decentralized applications use Next.js as the frontend and sometimes as the backend-for-frontend layer. In real audits, issues often span both sides: smart contracts and the web stack that exposes them.
This article focuses on the web execution side of that reality, not on-chain cryptography. If you are only interested in protocol-level or cryptographic audits, we publish separate articles that focus specifically on those topics.
The point here is that compromises do not respect category boundaries. They usually start at the web layer and move inward.
Out of curiosity, in your experience, do you usually see real-world compromises starting at the contract layer itself, or at the surrounding web and infrastructure layer that interfaces with it?
Just to address the “AI-generated” point directly:
This isn’t something you can realistically get out of an LLM by prompting it....
If you ask an AI to write about Next.js RCE, it will stay abstract, high-level, and defensive by default. It will avoid concrete execution paths, real integration details, or examples that could be interpreted as enabling exploitation — because that crosses into dual-use content.
This article deliberately goes further than that line: it includes real execution ordering, concrete framework behaviors, code-level examples, deployment patterns, and operational comparisons drawn from incident analysis. That’s exactly the kind of specificity automated filters tend to suppress or generalize away.
It’s still non-procedural on purpose — no payloads or step-by-step exploitation - but it’s not “AI vague” either. The detail is there so defenders can reason about where execution and observability actually break down.
Whether that level of detail is useful is subjective, but the reason it reads differently is because it’s grounded in real systems and real failure modes, not generated summaries.
To be clear, I’m not claiming this is some universal or inevitable failure mode, or that everyone running Next.js is compromised.
Every system has strengths and weaknesses. This is just one area where the tradeoffs aren’t always modeled correctly.
I don’t know what your setup looks like, how you deploy, or what your threat model is. You might already be accounting for this, or it might not matter for your use case. That’s fine.
The only point I’m making is that in modern SSR frameworks, execution can happen earlier than many teams expect — during deserialization, hydration, or framework setup — and when failures occur there, the signals look very different:
generic 500
no route handler invoked
no app logs
no auth context
That’s meaningfully different from traditional request-handling bugs that fail inside application control flow and leave traces people are used to seeing.
I’m not trying to persuade anyone or sell a solution. If you don’t find this relevant, you can safely ignore it.
But if you do run SSR in a security-sensitive environment, it doesn’t hurt to double-check where you believe the trust boundary actually starts — because in some cases it starts earlier than the app code.
Modern Next.js apps execute attacker-controlled input earlier than most teams realize — during framework deserialization, hydration, and Server Action resolution, often before application logging, validation, or auth hooks run.
In several real-world RCE investigations and red-team simulations, repeated 500 Internal Server Errors weren’t “noise” but early execution signals used by attackers to map execution boundaries and refine payloads. In some cases, the last observable 500 occurred right before stable code execution was achieved.
This write-up breaks down:
why deserialization in Next.js is part of execution, not preparation
how silent 500s can indicate pre-handler execution paths
why WAFs and app-level logs frequently miss this class of attacks
where the real attack surfaces live (middleware, RSC, Server Actions, custom servers)
Posting to get feedback from people who’ve seen or investigated similar SSR/RCE behavior in production.