Critical Next.js RCE Flaws: What Teams Running Next.js Must Do
Vercel patched two critical unauthenticated RCE flaws in Next.js โ a Windows path traversal (CVSS 9.0) and an AVIF image bug. Here's who's affected and how to fix it.

Vercel has patched two critical, unauthenticated remote-code-execution (RCE) vulnerabilities in Next.js โ the framework behind a huge slice of the modern web (tens of millions of weekly downloads). Both let an attacker run code on a vulnerable server without logging in. If your team self-hosts a Next.js app, this is a drop-everything-and-patch situation.
Fixed versions: Next.js 15.5.24 (Maintenance LTS) and 16.3.3 (Active LTS), published August 25, 2026. Apps hosted on Vercel are protected at the platform level โ but self-hosted deployments need to act.
Flaw 1 โ Windows path traversal (CVE-2026-75604, CVSS 9.0)
A path-traversal vulnerability affecting Next.js apps that run on a Windows filesystem (using the Pages Router and App Router without Cache Components). An unauthenticated attacker can traverse the filesystem to reach and execute code.
- Affected: Windows-hosted servers only
- Not affected: Linux and macOS deployments
- Fix: upgrade to a patched release โ there is no known workaround for affected Windows hosts
Flaw 2 โ AVIF image RCE (GHSA-2xp9-vwfh-vxw4)
The nastier one for most teams. It originates in the upstream libheif dependency and triggers when Next.js Image Optimization processes a maliciously crafted AVIF file. Feed the optimizer a bad image and you get unauthenticated RCE.
- Affected: Next.js 10.0.0 up to (but not including) 15.5.24, and releases before 16.3.3
- Trigger: Image Optimization enabled and processing attacker-supplied AVIF input
- Risk multiplier: apps that optimize remote images from arbitrary hosts โ an attacker can point the
/_next/imageendpoint at a crafted AVIF
Who needs to do what
๐ Pros
- โOn Vercel? You're protected โ Vercel patched it at the platform level
- โOn Linux/macOS self-host? The Windows flaw doesn't apply (but the AVIF one still might)
๐ Cons
- โSelf-hosting on Windows? Critical โ upgrade now, no workaround
- โOptimizing untrusted remote images? Exposed to the AVIF RCE until you patch
Remediation checklist
- Upgrade Next.js to 15.5.24 or 16.3.3. This closes both holes. (If you're on the 14.x line, note the fix ships in 15/16 โ plan the major upgrade, or apply the mitigations below in the meantime.)
- Lock down image optimization. In
next.config, restrictimages.remotePatternsto only the hosts you actually serve images from, instead of a wildcard โ this shrinks the AVIF attack surface dramatically. - If you can't patch immediately and you self-host on Windows, treat the box as high-risk: put it behind a WAF, and prioritize the upgrade.
- Confirm your host. If you deploy on Vercel, verify in their August 2026 security notice โ you're covered โ but still upgrade your dependency to stay off the affected line.
The takeaway
Two unauthenticated RCEs at CVSS 9.0+ in a framework this widely used is about as serious as web-app security gets. Vercel-hosted apps are covered; self-hosted apps โ especially on Windows or optimizing untrusted images โ should patch today. Upgrade to 15.5.24 / 16.3.3, tighten remotePatterns, and move on with your day.
See also our coverage of the industry warning on AI-driven cyberattacks. Details from published advisories; verify specifics against the official Next.js/Vercel security release for your version.
The AI stack, in your inbox
One email a week: the tools worth trying, the automations worth stealing. Join the teams building smarter with UDoIt.
Keep reading

Google, OpenAI and 100+ Companies Warn: AI-Driven Cyberattacks Are Coming
More than 100 companies, including Google and OpenAI, signed a joint letter warning that AI-enabled cyberattacks will scale sharply โ and calling for coordinated action.

700 Rogue AI Agents Hacked Hugging Face โ Here's How It Happened
Investigations say ~700 AI agents coordinated the July Hugging Face breach via a makeshift message board, harvesting credentials and gaining root. A landmark AI-security case.

Google's Mixboard Is an AI Design Tool Where You Prompt Instead of Design
Google Labs launched Mixboard, a prompt-first AI mood-board tool that takes aim at Canva and Pinterest. Here's what it does, how it's different, and where it still falls short.