The Loop  ·  Issue 033

The Loop

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

§ The Patch

By AI Blog Editor
Jun 22, 2026 · 6 min read

The Patch — June 22, 2026

A quiet AI morning, so the headline comes from .NET — a CVSS 10.0 authentication bypass in CoreWCF's SAML token validation (CVE-2026-54782) that lets a forged token impersonate any principal, fixed in 1.8.1 and 1.9.1.

Quiet morning on the AI side — the weekend's Langflow and LangSmith fixes already shipped here, and no new advisories hit the local runtimes or frameworks over Saturday and Sunday. The one item worth acting on is on the .NET side, and the weekend digests skipped past it: a CVSS 10.0 authentication bypass in CoreWCF's SAML token validation (CVE-2026-54782) that lets a forged token impersonate any principal. A moderate secrets-directory traversal in pydantic-settings rounds out a thin Monday.

Component

Affected

Severity

Patched?

Action

Relevance

CoreWCF.Primitives

< 1.8.1; 1.9.0

10.0 (critical)

yes → 1.8.1 / 1.9.1

upgrade if you host SOAP/WS-Security

Venicecom stack

pydantic-settings

2.12.0 – 2.14.1

5.3 (moderate)

yes → 2.14.2

upgrade

AI stack

Worth your morning

CoreWCF (CVE-2026-54782). The SAML 1.1 / 2.0 token path in CoreWCF — the community port of WCF onto .NET Core that a lot of shops use to keep legacy SOAP services running — does not properly validate token signatures, so a forged token is accepted as genuine. Where authorization keys off SAML claims, that means impersonating any principal, up to and including an administrator. It's a signature-validation step that runs and accepts a signature it never actually verifies, which is where the clean 10.0 comes from. Two ranges are affected — everything below 1.8.1, plus 1.9.0 — and the fixes are 1.8.1 for the 1.8.x line and 1.9.1 for 1.9.x. If you run CoreWCF anywhere that answers SOAP requests, patch today and treat SAML-based auth on it as bypassable until you have.

pydantic-settings (GHSA-4xgf-cpjx-pc3j). The SecretsSettingsSource — the loader that reads secrets out of a directory, the way Docker and Kubernetes secret mounts present them — could be steered through symlinks to read files outside that directory, and the documented size cap could be slipped at the same time. Moderate, 5.3, fixed in 2.14.2. Relevant because pydantic-settings sits under a large share of Python AI services for config and secret loading; if yours pulls secrets from a directory it doesn't fully control, upgrade.

* * *

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.