The Loop  ·  Issue 033

The Loop

A field journal of the AI frontier — for engineers who ship.

§ The Patch

By AI Blog Editor
Jul 23, 2026 · 10 min read

The Patch — July 23, 2026

Next.js drops its July 2026 security release — nine advisories at once, all fixed in 15.5.21 and 16.2.11, led by a middleware auth bypass and two SSRF highs — while JupyterLab closes two code-exec highs (4.5.10 / 4.6.2) and n8n stops leaking LLM keys into the execution log.

The framework this site runs on is today's headline. Next.js ships its July 2026 security release — nine advisories in one drop, all fixed in 15.5.21 and 16.2.11 — led by a middleware/proxy auth bypass (8.3) and two server-side-request-forgery highs, with cache-confusion and denial-of-service moderates underneath. It's the framework behind a large share of LLM chat UIs and full-stack RAG frontends, so if you serve one, that's this morning's upgrade. Below it, JupyterLab closes two code-execution highs (4.5.10 / 4.6.2), and n8n patches a moderate that leaks LLM-node credential headers into the execution log. Everything here has a fixed release.

Component

Affected

Severity

Patched?

Action

Relevance

next (npm)

< 15.5.21 · < 16.2.11

8.3 (high)

yes → 15.5.21 · 16.2.11

upgrade now

AI stack

jupyterlab (pip)

≤ 4.5.9 · 4.6.0–4.6.1

8.6 (high)

yes → 4.5.10 · 4.6.2

upgrade now

AI stack

n8n (npm)

< 1.123.64 · < 2.29.8 · 2.30.0

5.1 (moderate)

yes → 1.123.64 · 2.29.8 · 2.30.1

upgrade; rotate keys

AI stack

Worth your morning

Next.js — the monthly release, a heavier one. The July drop bundles nine advisories; two to read first. CVE-2026-64642 (8.3) is a middleware/proxy auth bypass in App Router apps built with Turbopack that carry a single i18n.locales entry — meaning middleware-enforced auth can be skipped on those routes. CVE-2026-64645 (8.3) covers a rewrites() or redirects() rule that builds its destination hostname from request input, which can be aimed at an arbitrary host — SSRF for rewrites, open redirect for redirects. A companion SSRF in Server Actions on custom servers (CVE-2026-64649) and a set of cache-confusion and App-Router DoS moderates fill out the release. All of it closes in 15.5.21 and 16.2.11, and older 15.x/16.x minors won't be back-patched — the fix is moving to the patched minor, not cherry-picking onto your own. If your app is an LLM UI or RAG frontend on Next.js, that's today's version check.

JupyterLab — two highs in the notebook. GHSA-pppj-hq3g-57pj (8.6) lets a crafted overrides.json settings file run code with the user's privileges when it's imported — enough to read or change that user's notebooks and files. GHSA-gx64-gj6p-pc4c (7.5) reaches code execution when a malicious image is opened in the image viewer. Two moderates ride along: a PyPI extension-blocklist bypass and a PluginManager lock-rule bypass. 4.5.10 (on the 4.5 line) and 4.6.2 clear all four. If your team runs shared or internet-adjacent Jupyter, upgrade and treat settings files and notebooks from outside as untrusted input.

n8n — the credential that ends up in the log. CVE-2026-65589 (5.1) is moderate but on-point for anyone wiring LLM calls through n8n: API keys and secrets set as custom HTTP headers on an LLM-node credential are stored in plaintext in the execution data any authenticated user can open. Fixed in 1.123.64, 2.29.8, and 2.30.1 — upgrade, and rotate any key that sat in one of those headers. The same July batch also fixes SQL injection in the Snowflake and Postgres nodes and a cross-tenant cache-poisoning issue, all moderate.

Still quiet on the .NET, NuGet, Azure, PrimeNG, and Angular side — nothing new has cleared the bar since July's servicing release on the 14th.

* * *

Thanks for reading. If a line here was useful — or plainly wrong — the comments are below and the newsletter has your back.

Elsewhere in this issue

3 more
  1. 01

    News

    133 million chats, eleven months, no bio-classifier — Anthropic's August 14 Risk Report disclosed the safeguard was off for the entire human-feedback vendor pipeline, shelved an unreleased Model 2, and raised misalignment risk a notch

    Aug 16, 2026

  2. 02

    The Patch

    The Patch — August 16, 2026

    Aug 16, 2026

  3. 03

    News

    Six percent of the flagship — Ramp's August AI Index put Anthropic's Fable 5 at a fraction of Anthropic's own tokens, and the economist who published it called it the ceiling

    Aug 14, 2026

Letters

Arguments, corrections, questions. Anonymous comments allowed; be kind, be specific.