Today on The Staff Safety Desk: critical vulnerabilities hitting Django serializers and administrative views, an arbitrary code execution flaw in PostgreSQL's refint extension, and empty default secrets in webhook payment integrations.
Advisory CVE-2026-112358 identifies a server-side request forgery flaw in TiBillet Lespass APIv1 up to commit 10e8790. Located in the get_img_from_url function inside ApiBillet/serializers.py, the endpoint fetches user-supplied img_url parameters without destination IP validation or protocol timeouts. The maintainers confirmed APIv1 endpoints will be deprecated in the next major release.
Why it matters
Allowing Django Rest Framework serializers to pull remote assets directly from user input lets remote attackers query internal cloud metadata services and bypass private subnet boundaries.
Security advisory CVE-2026-112289 discloses a stored cross-site scripting vulnerability in Koalixcrm through version 1.14.0. Malicious task titles supplied by staff users are rendered unescaped via Task.link_to_task() in the Django Admin interface, executing arbitrary client-side scripts when administrators view changelists or change forms. No official upstream fix has been released yet.
Why it matters
If your application relies on raw administrative helper methods that mark string outputs safe without explicit escaping, authenticated staff users can steal admin session tokens or trigger unauthenticated background queries.
The push to validate webhook HMAC signatures against raw request byte buffers—a pattern we tracked recently to prevent JSON re-serialization failures—has collided with autonomous coding agents. A new breakdown of defense-in-depth webhook ingestion demonstrates how these constant-time HMAC comparisons broke in staging when LLM code generators injected conversational markdown preambles and non-ASCII characters directly into the raw backend source files.
Why it matters
Unsanitized AI code streams bypass syntax compilation gates when models emit conversational markdown around Python functions, breaking downstream build pipelines despite correct logic.
Following the Lamplit Labs data we covered yesterday showing that mechanically verifiable AGENTS.md rules cut AI pull request rework to 17%, a new architectural guide demonstrates how to practically implement these zero-memory workflows. To replace unstructured text lists, the author recommends capping rule files under 200 lines, providing path-based directory maps, defining absolute forbidden patterns, and pinning rules directly to deterministic linter checks.
Why it matters
Treating agent instruction files as unstructured prompt dumps bloats context windows and leads to hallucinated abstractions, whereas backing rules with automated CI gates forces deterministic compliance.
Red Hat and PostgreSQL downstream maintainers issued advisories for CVE-2026-14671, a CVSS 8.8 type confusion vulnerability in the PostgreSQL refint contrib module. The flaw allows authenticated database users capable of running queries to execute arbitrary operating system commands as the postgres system user. Updated packages have been pushed across supported OS repositories.
Why it matters
Legacy contrib extensions loaded into production database clusters bypass application-level authorization controls, converting standard SQL access into full underlying host compromise.
Security disclosure CVE-2026-103311 details a vulnerability in the Coinbase Commerce Payment Gateway plugin through version 1.4.1. The inbound webhook handler defaults its shared verification secret to an empty string when unconfigured, allowing unauthenticated attackers to sign valid payment payloads and mark unpaid orders as completed.
Why it matters
Integrations that treat missing or empty secret environment variables as valid signatures allow external actors to forge payment completion callbacks without spending funds.
Unvalidated Outbound Fetching in API Serializers Exposes Internal Microservices Recent CVEs in Django-based APIs show a recurring vulnerability pattern where serializers accept external image URLs without validating destination IP addresses or verifying redirect hops. When combined with synchronous HTTP calls inside request handlers, attackers can trigger blind SSRF against cloud metadata endpoints.
Unconfigured Integration Defaults Fail Open Instead of Failing Closed Payment integration disclosures reveal that webhook handlers routinely process callbacks with uninitialized shared secrets rather than rejecting missing tokens. When libraries default to empty strings, any unauthenticated attacker can forge payment status notifications.
Legacy Contrib Modules Injected Into Production Databases Create Unmonitored RCE Surfaces High-severity vulnerabilities in legacy PostgreSQL C-extensions demonstrate that database security extends far beyond standard core queries. Contrib modules handling referential integrity or custom types often bypass standard application-level access controls.
What to Expect
2026-11-12—PostgreSQL 14 reaches official community End-of-Life (EOL) and stops receiving security patches.
2026-11-19—PostgreSQL 19 general availability release target following code-review maintenance delays.
2026-12-01—Vietnam Thong tu 60/2026/TT-BKHCN standardizing public administrative technical rules takes effect.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
337
📖
Read in full
Every article opened, read, and evaluated
97
⭐
Published today
Ranked by importance and verified across sources
6
— The Staff Safety Desk
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste