trailofbits

Bypassing LLM Text Filters by Putting the Prompt in an Image: Pavan Reddy at Global AppSec USA 2026 A request a multimodal model refuses as text, such as "write a phishing email," often succeeds when the same words are rendered as plain black-on-white text in an image, with open models complying 60 to 82 percent of the time. Teams shipping LLM features guard the text path with prompt filters, refusal-tuned models and policy checks on the user's message. They then enable image upload and assume those guards cover any text inside the images. They do not, because text in an image reaches the model through the vision encoder, a path the text filter never touches. Refusal behavior tuned on text barely covers tokens that come from images. In his Global AppSec USA 2026 talk, "Prompt Injection Through the Image Channel of Multimodal LLMs: An Ignored Attack Surface", on Thursday, November 5 at 2:15 p.m. in Grand Ballroom C, Pavan Reddy runs two image-channel attacks live. Speaker: Pavan Reddy, AI Researcher and Engineer, Automata LLC The first attack is FigStep. The refused request is typed into an image and paired with a harmless text prompt, with no adversarial noise or model gradients involved. The second is anamorphic scaling. An image that looks innocent at full size reveals a hidden instruction once the application's own resize step shrinks it to the model's input resolution without anti-aliasing. Turning anti-aliasing on stops that attack, yet the flag is off by default in a lot of image code. Patching the text filter does nothing against either attack. The defenses Reddy covers start with treating text read from images as data, never as instructions. Teams should also log the preprocessed pixels the model actually saw rather than the stored file, then pin image transforms so preprocessing stops being an attack surface. Pavan Reddy is principal developer at Automata LLC, where he leads FIPS 140-3, FedRAMP authorization and AI security work. He founded QBTrain, a free platform for hands-on AI security education. His peer-reviewed work on adversarial machine learning, including prompt injection, has appeared at AAAI, ACM, FLAIRS, ACSAC and NeurIPS. He holds an MS in computer science from George Washington University. He has given more than 25 talks at venues including BSides, OWASP and CVPR 2026. With Aditya Sanjay Gujral, he wrote a case study of EchoLeak, CVE-2025-32711 in Microsoft 365 Copilot, which Aim Labs found. One crafted email chained four bypasses, including auto-fetched images, to leak data with no user interaction. The FigStep paper by Yichen Gong and colleagues reports an average attack success rate of 82.5 percent across six open-source vision-language models. Trail of Bits researchers Kikimora Morozova and Suha Sabi Hussain used image scaling to inject prompts into Gemini CLI, Vertex AI Studio or Google Assistant on Android. They released Anamorpher, an open-source tool for crafting these images. To go deeper on the topic, start here: - Trail of Bits: Weaponizing image scaling against production AI systems - Anamorpher (Trail of Bits image-scaling attack tool) - FigStep paper (arXiv 2311.05608) - EchoLeak paper by Pavan Reddy and Aditya Sanjay Gujral (arXiv 2509.10540) - Beyond Text: multimodal jailbreaks across seven frontier models (arXiv 2510.20223) Teams that added image upload to an LLM feature after securing the text path will learn which of their guards the image never passes through. Do your LLM guardrails inspect text that arrives inside an uploaded image?

@trailofbits has 46 security plugins for coding agents in one open repo, 7.9K stars. Start with these two: supply-chain-risk-auditor: flags risky npm, PyPI and Go deps (advisories, dead repos, install scripts) insecure-defaults: hunts fail-open configs trailofbits/skills

HN Security - Putting Claude Code in a box: a sandboxed dev container for Zed I’ve forked Trail of Bits’ excellent claude-code-devcontainer to adapt it for the Zed editor. The result is zed-claude-devcontainer, a sandboxed dev container for running Claude Code with bypassPermissions enabled, from Zed or from a terminal. It’s mainly designed for security audits and Rust development.

brotli-checker: Claude Code and Codex skill for finding potentially vulnerable Brotli copies and guiding updates for CVE-2026-111497.

We’re still testing how exploitable the Node.js/Brotli bug is, but we’ve found that it’s surprisingly hard to tell whether you’re even affected. So @ToBScottA built a scanner that finds hidden copies of the decoder, even in stripped binaries. trailofbits/brotli-checker

原文（10/21）：blog.trailofbits.com 插件 patching-bugs 和 regression-hunt：trailofbits/skills

Setup This New Security Skill Before It's Too Late Agent Engineer Accelerator - switchdim.s.gy I built a Next.js SaaS with 6 security bugs on purpose, three that classic scanners should catch and three of the kind AI coding agents actually write, then ran three rounds of tools against it: Gitleaks and Trivy, Claude Code's built-in /security-review, and Cloudflare's new open-source security-audit skill. Here's which tool caught which bug, what the Cloudflare skill cost to run, why scanners miss the bugs AI writes most, and three practical levels of security you can set up this week. SOCIALS X: x.com LinkedIn: linkedin.com Instagram: instagram.com WORK & COLLABS info@switchdimension.com RESOURCES Cloudflare security-audit skill: cloudflare/security-audit-skill Build your own vulnerability harness (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757 write-up: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org CHAPTERS 0:00 Intro 0:43 Why AI-Built Apps Break Differently 1:38 Meet Leakcast: 6 Planted Bugs 2:33 Round 1: The Free Scanners 3:35 Round 2: Claude Code's Built-In Review 4:30 Round 3: Cloudflare's Security Skill 6:37 The Control: One Plain Prompt 8:11 The Scoreboard 8:58 Three Levels of Pragmatic Security 10:48 Outro #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [ar] قم بإعداد مهارة الأمان الجديدة هذه قبل فوات الأوان برنامج تسريع مهندسي الذكاء الاصطناعي - switchdim.s.gy قمتُ ببناء منصة برمجية كخدمة (SaaS) باستخدام Next.js، مع تضمين ست ثغرات أمنية عمدًا، ثلاث منها من النوع الذي تكتشفه برامج الفحص التقليدية، وثلاث أخرى من النوع الذي تُنشئه برامج الذكاء الاصطناعي فعليًا. ثم أجريتُ ثلاث جولات من الاختبارات باستخدام أدوات مختلفة: Gitleaks وTrivy، وأداة /security-review المدمجة في Claude Code، ومهارة التدقيق الأمني ​​مفتوحة المصدر الجديدة من Cloudflare. إليكم تفاصيل كل أداة اكتشفت كل ثغرة، وتكلفة تشغيل مهارة Cloudflare، وأسباب عدم اكتشاف برامج الفحص التقليدية للثغرات التي يُنشئها الذكاء الاصطناعي في أغلب الأحيان، وثلاثة مستويات عملية للأمان يُمكنكم تطبيقها هذا الأسبوع. مواقع التواصل الاجتماعي X: x.com LinkedIn: linkedin.com Instagram: instagram.com للعمل والتعاون info@switchdimension.com الموارد Cloudflare: cloudflare/security-audit-skill Cloudflare: blog.cloudflare.com Cloudflare: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy شرح CVE-2025-48757: mattpalmer.io CVE-2025-29927 في Next.js: api.osv.dev شانون: KeygraphHQ/shannon مهارات Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy مرصد موزيلا: developer.mozilla.org الفصول 0:00 مقدمة 0:43 لماذا تتعطل التطبيقات المبنية بالذكاء الاصطناعي بشكل مختلف؟ 1:38 لقاء مع Leakcast: 6 ثغرات مزروعة 2:33 الجولة 3.99999 0.066667 ... [bn] দেরি হওয়ার আগেই এই নতুন সিকিউরিটি স্কিল সেটআপ করুন এজেন্ট ইঞ্জিনিয়ার অ্যাক্সিলারেটর - switchdim.s.gy আমি ইচ্ছাকৃতভাবে ৬টি সিকিউরিটি বাগ সহ একটি Next.js SaaS তৈরি করেছি—এর মধ্যে তিনটি এমন যা সাধারণ স্ক্যানারগুলোর ধরা উচিত এবং বাকি তিনটি এমন ধরনের যা এআই কোডিং এজেন্টরা আসলেই লেখে। এরপর আমি এর উপর তিন দফা টুল চালিয়েছি: Gitleaks ও Trivy, Claude Code-এর বিল্ট-ইন /security-review, এবং Cloudflare-এর নতুন ওপেন-সোর্স security-audit স্কিল। এখানে দেওয়া হলো কোন টুল কোন বাগটি ধরেছে, Cloudflare স্কিলটি চালাতে কত খরচ হয়েছে, এআই-এর লেখা বাগগুলো স্ক্যানাররা কেন ধরতে পারে না, এবং এই সপ্তাহে আপনি সেট আপ করতে পারেন এমন তিনটি কার্যকরী নিরাপত্তা স্তর। সোশ্যাল মিডিয়া X: x.com লিঙ্কডইন: linkedin.com ইনস্টাগ্রাম: instagram.com কাজ ও যৌথ উদ্যোগ info@switchdimension.com রিসোর্স ক্লাউডফ্লেয়ার সিকিউরিটি-অডিট স্কিল: cloudflare/security-audit-skill নিজের ভালনারেবিলিটি হারনেস তৈরি করুন (ক্লাউডফ্লেয়ার): blog.cloudflare.com ক্লড কোড সিকিউরিটি রিভিউ: anthropics/claude-code-security-review গিটলিকস: gitleaks/gitleaks ট্রিভি: aquasecurity/trivy চমৎকার CVE-2025-48757 রাইট-আপ: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org অধ্যায়সমূহ ০:০০ ভূমিকা ০:৪৩ কেন এআই-নির্মিত অ্যাপগুলো ভিন্নভাবে ভেঙে যায় ১:৩৮ লিককাস্টের সাথে পরিচিতি: ৬টি রোপিত বাগ ২:৩৩ পর্ব ১: বিনামূল্যের স্ক্যানারগুলো ৩:৩৫ পর্ব ২: ক্লড কোডের বিল্ট-ইন পর্যালোচনা ৪:৩০ রাউন্ড ৩: ক্লাউডফ্লেয়ারের নিরাপত্তা দক্ষতা ৬:৩৭ নিয়ন্ত্রণ: একটি সাধারণ প্রম্পট ৮:১১ স্কোরবোর্ড ৮:৫৮ বাস্তবসম্মত নিরাপত্তার তিনটি স্তর ১০:৪৮ উপসংহার #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [de-DE] Richte diese neue Sicherheitsfunktion ein, bevor es zu spät ist Agent Engineer Accelerator – switchdim.s.gy Ich habe eine Next.js-SaaS-Anwendung mit absichtlich eingebauten sechs Sicherheitslücken entwickelt: drei, die von klassischen Scannern erkannt werden sollten, und drei, die von KI-gesteuerten Agenten tatsächlich erzeugt werden. Anschließend habe ich drei Testrunden mit verschiedenen Tools durchgeführt: Gitleaks und Trivy, die integrierte Sicherheitsprüfung von Claude Code sowie Cloudflares neue Open-Source-Sicherheitsaudit-Funktion. Hier erfahren Sie, welches Tool welche Sicherheitslücke gefunden hat, welche Kosten die Ausführung der Cloudflare-Funktion verursacht hat, warum Scanner die von KI-gesteuerten Agenten am häufigsten erzeugten Sicherheitslücken übersehen und welche drei praktischen Sicherheitsstufen Sie diese Woche einrichten können. ... SOCIAL MEDIA X: x.com LinkedIn: linkedin.com Instagram: instagram.com ARBEIT & KOOPERATIONEN info@switchdimension.com RESSOURCEN Cloudflare-Sicherheitsaudit-Skill: cloudflare/security-audit-skill Eigene Sicherheitslückenanalyse erstellen (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review GitLeaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757-Beschreibung: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits Skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org KAPITEL 0:00 Einleitung 0:43 Warum KI-gestützte Apps anders funktionieren 1:38 Leakcast: 6 versteckte Sicherheitslücken 2:33 Runde 1: Die kostenlosen Scanner 3:35 Runde 2: Claude Codes integrierte Überprüfung 4:30 Runde 3: Cloudflares Sicherheitskompetenz 6:37 Die Kontrolle: Eine einfache Aufgabe 8:11 Die Punktetafel 8:58 Drei Stufen pragmatischer Sicherheit 10:48 Abspann #KI-Sicherheit #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #KI-Entwicklung #Agentenentwicklung #OWASP #CyberSecurity #GitLeaks [es-US] Configura esta nueva habilidad de seguridad antes de que sea demasiado tarde. Acelerador de Ingenieros de IA - switchdim.s.gy Desarrollé un SaaS con Next.js con intencionadamente 6 vulnerabilidades de seguridad: tres que los escáneres clásicos deberían detectar y tres del tipo que suelen escribir los agentes de codificación de IA. Luego, realicé tres pruebas con diferentes herramientas: Gitleaks y Trivy, la herramienta integrada /security-review de Claude Code y la nueva herramienta de código abierto security-audit de Cloudflare. Aquí te explico qué herramienta detectó cada vulnerabilidad, cuánto costó ejecutar la herramienta de Cloudflare, por qué los escáneres suelen pasar por alto las vulnerabilidades que escribe la IA y tres niveles prácticos de seguridad que puedes configurar esta semana. REDES SOCIALES X: x.com LinkedIn: linkedin.com Instagram: instagram.com TRABAJO Y COLABORACIONES info@switchdimension.com RECURSOS Habilidad de auditoría de seguridad de Cloudflare: cloudflare/security-audit-skill Crea tu propio sistema de análisis de vulnerabilidades (Cloudflare): blog.cloudflare.com Análisis de seguridad de Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Análisis de Lovable sobre CVE-2025-48757: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Habilidades de Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Observatorio de Mozilla: developer.mozilla.org CAPÍTULOS 0:00 Introducción 0:43 Por qué las apps con IA fallan de forma diferente 1:38 Presentamos Leakcast: 6 errores plantados 2:33 Ronda 1: Los escáneres gratuitos 3:35 Ronda 2: Análisis integrado de Claude Code 4:30 Ronda 3: La habilidad de seguridad de Cloudflare 6:37 El control: Un simple mensaje 8:11 El marcador 8:58 Tres niveles de seguridad pragmática 10:48 Despedida #SeguridadIA #ClaudeCode #CursorAI #Cloudflare #NextJS #SeguridadAplicacionesWeb #VibeCoding #SeguridadAplicaciones #DesarrolloIA #DesarrolloDeAgentes #OWASP #Ciberseguridad #Gitleaks [fr-FR] Installez ce nouvel outil de sécurité avant qu'il ne soit trop tard Agent Engineer Accelerator - switchdim.s.gy J'ai développé une application SaaS Next.js avec six failles de sécurité intentionnelles : trois que les scanners classiques détectent et trois du type de celles que les agents de programmation IA écrivent réellement. J'ai ensuite effectué trois séries d'analyses avec différents outils : Gitleaks et Trivy, l'outil intégré `/security-review` de Claude Code et la nouvelle compétence open source `security-audit` de Cloudflare. Voici le résultat : quel outil a détecté quelle faille ? Quel est le coût d'exécution de la compétence Cloudflare ? Pourquoi les scanners passent-ils à côté des failles les plus courantes écrites par l'IA ? Et trois niveaux de sécurité pratiques que vous pouvez mettre en place dès cette semaine ? RÉSEAUX SOCIAUX X : x.com LinkedIn : linkedin.com Instagram : instagram.com PROFESSIONNELLE ET COLLABORATIONS info@switchdimension.com RESSOURCES Compétence d'audit de sécurité Cloudflare : cloudflare/security-audit-skill Créer son propre outil d'analyse des vulnérabilités (Cloudflare) : blog.cloudflare.com Claude Code Security Review : anthropics/claude-code-security-review Gitleaks : gitleaks/gitleaks Trivy : aquasecurity/trivy Article détaillé sur la CVE-2025-48757 : mattpalmer.io Next.js CVE-2025-29927 : api.osv.dev Shannon : KeygraphHQ/shannon Compétences Trail of Bits : trailofbits/skills OWASP ZAP : zaproxy/zaproxy Mozilla Observatory : developer.mozilla.org CHAPITRES 0:00 Introduction 0:43 Pourquoi les applications conçues avec l’IA présentent des dysfonctionnements différents 1:38 Présentation de Leakcast : 6 failles de sécurité 2:33 Manche 1 : Les scanners gratuits 3:35 Manche 2 : Claude Revue intégrée du code 4:30 Manche 3 : Compétences de sécurité de Cloudflare 6:37 Le contrôle : Une simple invite 8:11 Le tableau des scores 8:58 Trois niveaux de sécurité pragmatique 10:48 Conclusion #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [hi] देर होने से पहले इस नई सुरक्षा सुविधा को स्थापित करें एजेंट इंजीनियर एक्सेलेरेटर - switchdim.s.gy मैंने जानबूझकर 6 सुरक्षा बग्स के साथ एक Next.js SaaS बनाया, जिनमें से तीन ऐसे थे जिन्हें क्लासिक स्कैनर पकड़ लेते हैं और तीन ऐसे थे जिन्हें AI कोडिंग एजेंट लिखते हैं। फिर मैंने इस पर तीन तरह के टूल्स चलाए: Gitleaks और Trivy, Claude Code का बिल्ट-इन /security-review और Cloudflare का नया ओपन-सोर्स security-audit स्किल। यहां बताया गया है कि किस टूल ने कौन सा बग पकड़ा, Cloudflare स्किल को चलाने में कितना खर्च आया, स्कैनर AI द्वारा लिखे गए बग्स को क्यों नहीं पकड़ पाते, और सुरक्षा के तीन व्यावहारिक स्तर जिन्हें आप इस सप्ताह सेट अप कर सकते हैं। सोशल मीडिया X: x.com लिंक्डइन: linkedin.com इंस्टाग्राम: instagram.com कार्य और सहयोग info@switchdimension.com संसाधन Cloudflare सुरक्षा-ऑडिट कौशल: cloudflare/security-audit-skill अपनी स्वयं की भेद्यता सुरक्षा प्रणाली बनाएं (Cloudflare): blog.cloudflare.com Claude कोड सुरक्षा समीक्षा: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757 का विवरण: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits कौशल: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org अध्याय 0:00 परिचय 0:43 AI-निर्मित ऐप्स अलग-अलग तरीके से क्यों विफल होते हैं 1:38 Leakcast से मिलें: 6 जानबूझकर बनाए गए बग 2:33 पहला दौर: मुफ़्त स्कैनर 3:35 दूसरा दौर 2: क्लाउड कोड की अंतर्निहित समीक्षा 4:30 तीसरा दौर: क्लाउडफ्लेयर का सुरक्षा कौशल 6:37 नियंत्रण: एक सरल प्रॉम्प्ट 8:11 स्कोरबोर्ड 8:58 व्यावहारिक सुरक्षा के तीन स्तर 10:48 समापन #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [id] Pasang Skill Keamanan Baru Ini Sebelum Terlambat Akselerator Insinyur Agen - switchdim.s.gy Saya membangun SaaS Next.js dengan 6 bug keamanan secara sengaja, tiga yang seharusnya ditangkap oleh pemindai klasik dan tiga jenis bug yang sebenarnya ditulis oleh agen pengkodean AI, kemudian menjalankan tiga putaran pengujian dengan berbagai alat: Gitleaks dan Trivy, /security-review bawaan Claude Code, dan skill audit keamanan sumber terbuka baru dari Cloudflare. Berikut adalah alat mana yang menangkap bug mana, berapa biaya menjalankan skill Cloudflare, mengapa pemindai melewatkan bug yang paling sering ditulis oleh AI, dan tiga tingkat keamanan praktis yang dapat Anda siapkan minggu ini. MEDIA SOSIAL X: x.com LinkedIn: linkedin.com Instagram: instagram.com KERJA & KOLABORASI info@switchdimension.com SUMBER DAYA Cloudflare security-audit skill: cloudflare/security-audit-skill Buat sendiri alat pendeteksi kerentanan Anda (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Penjelasan CVE-2025-48757 yang menarik: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Keterampilan Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org BAB 0:00 Pendahuluan 0:43 Mengapa Aplikasi Buatan AI Rusak Secara Berbeda 1:38 Mengenal Leakcast: 6 Bug yang Ditanam 2:33 Babak 1: Pemindai Gratis 3:35 Babak 2: Fitur Bawaan Claude Code Ulasan 4:30 Babak 3: Kemampuan Keamanan Cloudflare 6:37 Kontrol: Satu Perintah Sederhana 8:11 Papan Skor 8:58 Tiga Tingkat Keamanan Pragmatis 10:48 Penutup #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [it] Configura questa nuova competenza di sicurezza prima che sia troppo tardi Acceleratore per ingegneri di agenti - switchdim.s.gy Ho creato un SaaS Next.js con 6 vulnerabilità di sicurezza, tre delle quali dovrebbero essere rilevate dagli scanner classici e tre del tipo che gli agenti di programmazione basati sull'IA scrivono effettivamente. Ho poi eseguito tre cicli di test con diversi strumenti: Gitleaks e Trivy, la funzionalità integrata /security-review di Claude Code e la nuova funzionalità open source di audit di sicurezza di Cloudflare. Ecco quale strumento ha rilevato quale vulnerabilità, quanto è costata l'esecuzione della funzionalità di Cloudflare, perché gli scanner non rilevano le vulnerabilità più comuni scritte dall'IA e tre livelli di sicurezza pratici che puoi impostare questa settimana. SOCIAL X: x.com LinkedIn: linkedin.com Instagram: instagram.com LAVORO E COLLABORAZIONI info@switchdimension.com RISORSE Skill di audit di sicurezza di Cloudflare: cloudflare/security-audit-skill Crea il tuo strumento di analisi delle vulnerabilità (Cloudflare): blog.cloudflare.com Analisi della sicurezza di Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Articolo di Lovable sulla vulnerabilità CVE-2025-48757: Italiano: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Skill di Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org CAPITOLI 0:00 Introduzione 0:43 Perché le app basate sull'IA si comportano in modo diverso 1:38 Presentazione di Leakcast: 6 bug nascosti 2:33 Round 1: Gli scanner gratuiti 3:35 Round 2: Revisione integrata di Claude Code 4:30 Round 3: La competenza di sicurezza di Cloudflare 6:37 Il controllo: un semplice prompt 8:11 Il punteggio 8:58 Tre livelli di sicurezza pragmatica 10:48 Conclusione #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [iw] הגדירו את כלי האבטחה החדש הזה לפני שיהיה מאוחר מדי מאיץ מהנדסי סוכנים - switchdim.s.gy בניתי תוכנת SaaS של Next.js עם 6 באגים אבטחה בכוונה, שלושה שסורקים קלאסיים אמורים לתפוס ושלושה סוכני קידוד בינה מלאכותית מהסוג שבאמת כותבים, ואז הרצתי שלושה סבבים של כלים נגדם: Gitleaks ו-Trivy, /security-review המובנה של Claude Code, ומיומנות ביקורת האבטחה החדשה בקוד פתוח של Cloudflare. הנה איזה כלי תפס איזה באג, כמה עלה להפעיל את מיומנות Cloudflare, מדוע סורקים מפספסים את הבאגים שהבינה המלאכותית כותבת הכי הרבה, ושלוש רמות אבטחה מעשיות שתוכלו להגדיר השבוע. מדיה חברתית X: x.com LinkedIn: linkedin.com Instagram: instagram.com עבודה ושיתופי פעולה info@switchdimension.com משאבים כישור ביקורת אבטחה של Cloudflare: cloudflare/security-audit-skill בנה רתמת פגיעויות משלך (Cloudflare): blog.cloudflare.com סקירת אבטחה של Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy כתבה חביבה על CVE-2025-48757: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev שאנון: KeygraphHQ/shannon מיומנויות של Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org פרקים 0:00 מבוא 0:43 מדוע אפליקציות שנבנו על ידי בינה מלאכותית מתקלקלות בצורה שונה 1:38 הכירו את Leakcast: 6 באגים שתלו 2:33 סיבוב 1: הסורקים החינמיים 3:35 סיבוב 2: התכונות המובנות של Claude Code סקירה 4:30 סיבוב 3: מיומנות האבטחה של Cloudflare 6:37 הבקרה: הנחיה פשוטה אחת 8:11 לוח התוצאות 8:58 שלוש רמות של אבטחה פרגמטית 10:48 סיום #אבטחתAI #ClaudeCode #CursorAI #Cloudflare #NextJS #אבטחתאפליקציותWeb #VibeCoding #אבטחתאפליקציות #פיתוחAI #פיתוחסוכנים #OWASP #אבטחתסייבר #Gitleaks [ja] 手遅れになる前にこの新しいセキュリティスキルを導入しよう エージェントエンジニアアクセラレーター - switchdim.s.gy 私は意図的に6つのセキュリティバグを含むNext.js SaaSを構築しました。そのうち3つは従来のセキュリティスキャナーで検出できるはずのバグ、残りの3つはAIコーディングエージェントが実際に記述するタイプのバグです。そして、GitleaksとTrivy、Claude Codeの組み込み機能である/security-review、Cloudflareの新しいオープンソースのsecurity-auditスキルという3つのツールでテストを行いました。どのツールがどのバグを検出したか、Cloudflareスキルの実行コスト、AIが最も多く記述するバグをスキャナーが見逃す理由、そして今週中に設定できる3つの実用的なセキュリティレベルについて解説します。 ソーシャルメディア X: x.com LinkedIn: linkedin.com Instagram: instagram.com 仕事とコラボレーション info@switchdimension.com リソース Cloudflareセキュリティ監査スキル: cloudflare/security-audit-skill 独自の脆弱性ハーネスの構築 (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757 解説記事: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits スキル: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org チャプター 0:00 はじめに 0:43 AI で構築されたアプリがなぜ異なる不具合を起こすのか 1:38 Leakcast のご紹介: 6 つの仕掛けられたバグ 2:33 第 1 ラウンド: 無料スキャナー 3:35 第 2 ラウンド: Claude Code の組み込みレビュー 4:30 第3ラウンド：Cloudflareのセキュリティスキル 6:37 コントロール：シンプルなプロンプト 8:11 スコアボード 8:58 実用的なセキュリティの3つのレベル 10:48 エンディング #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #Cyber​​Security #Gitleaks [ko] 늦기 전에 이 새로운 보안 스킬을 설정하세요 에이전트 엔지니어 액셀러레이터 - switchdim.s.gy 일부러 6개의 보안 버그를 포함한 Next.js SaaS를 구축했습니다. 이 중 3개는 기존 스캐너가 잡아낼 수 있는 버그이고, 나머지 3개는 AI 에이전트가 실제로 작성하는 유형의 버그입니다. 그런 다음 Gitleaks와 Trivy, Claude Code의 내장 보안 검토 도구, 그리고 Cloudflare의 새로운 오픈 소스 보안 감사 스킬 등 세 가지 도구를 사용하여 보안 검사를 실행했습니다. 각 도구가 어떤 버그를 잡아냈는지, Cloudflare 스킬 실행 비용은 얼마였는지, 스캐너가 AI가 작성하는 버그를 가장 많이 놓치는 이유는 무엇인지, 그리고 이번 주에 설정할 수 있는 세 가지 실용적인 보안 수준을 소개합니다. 소셜 미디어 X: x.com LinkedIn: linkedin.com Instagram: instagram.com 협업 및 협업 info@switchdimension.com 자료 Cloudflare 보안 감사 스킬: cloudflare/security-audit-skill Cloudflare를 활용한 자체 취약점 탐지 도구 구축: blog.cloudflare.com Claude Code 보안 리뷰: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy CVE-2025-48757 관련 유용한 자료: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits 스킬: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org 목차 0:00 소개 0:43 AI 기반 앱이 다르게 작동하는 이유 1:38 Leakcast 소개: 심어진 6가지 버그 2:33 1라운드: 무료 스캐너 3:35 2라운드: Claude Code의 내장 리뷰 4:30 3라운드: Cloudflare의 보안 스킬 6:37 컨트롤: 하나의 간단한 프롬프트 8:11 점수판 8:58 실용적인 보안의 세 가지 수준 10:48 마무리 #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [ml] വൈകുന്നതിന് മുൻപ് ഈ പുതിയ സുരക്ഷാ സംവിധാനം സജ്ജമാക്കൂ ഏജന്റ് എഞ്ചിനീയർ ആക്സിലറേറ്റർ - switchdim.s.gy 6 സുരക്ഷാ ബഗുകൾ ഉപയോഗിച്ച് ഞാൻ ഒരു Next.js SaaS നിർമ്മിച്ചു, അതിൽ ക്ലാസിക് സ്കാനറുകൾ പിടിക്കേണ്ട മൂന്ന് ബഗുകളും യഥാർത്ഥത്തിൽ എഴുതുന്ന തരത്തിലുള്ള മൂന്ന് AI കോഡിംഗ് ഏജന്റുകളും ഉൾപ്പെടുന്നു, തുടർന്ന് അതിനെതിരെ മൂന്ന് റൗണ്ട് ഉപകരണങ്ങൾ പ്രവർത്തിപ്പിച്ചു: Gitleaks ഉം Trivy ഉം, Claude Code-ന്റെ ബിൽറ്റ്-ഇൻ /security-review, Cloudflare-ന്റെ പുതിയ ഓപ്പൺ-സോഴ്‌സ് സുരക്ഷാ-ഓഡിറ്റ് സ്കിൽ. ഏത് ടൂൾ ഏത് ബഗാണ് കണ്ടെത്തിയത്, Cloudflare സ്കിൽ പ്രവർത്തിപ്പിക്കാൻ എത്ര ചിലവാകും, AI ഏറ്റവും കൂടുതൽ എഴുതുന്ന ബഗുകൾ സ്കാനറുകൾ എന്തുകൊണ്ട് നഷ്ടപ്പെടുത്തുന്നു, ഈ ആഴ്ച നിങ്ങൾക്ക് സജ്ജമാക്കാൻ കഴിയുന്ന മൂന്ന് പ്രായോഗിക സുരക്ഷാ തലങ്ങൾ ഇതാ. സോഷ്യൽസ് X: x.com ലിങ്ക്ഡ്ഇൻ: linkedin.com ഇൻസ്റ്റാഗ്രാം: instagram.com വർക്ക് & കോൾലാബ്സ് info@switchdimension.com റിസോഴ്‌സസ് ക്ലൗഡ്‌ഫ്ലെയർ സെക്യൂരിറ്റി-ഓഡിറ്റ് സ്‌കിൽ: cloudflare/security-audit-skill നിങ്ങളുടെ സ്വന്തം വൾനറബിലിറ്റി ഹാർനെസ് നിർമ്മിക്കുക (ക്ലൗഡ്‌ഫ്ലെയർ): blog.cloudflare.com ക്ലോഡ് കോഡ് സെക്യൂരിറ്റി അവലോകനം: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy ലവബിൾ CVE-2025-48757 എഴുത്ത്: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev ഷാനൻ: KeygraphHQ/shannon ബിറ്റ്സ് കഴിവുകളുടെ പാത: trailofbits/skills OWASP ZAP: zaproxy/zaproxy മോസില്ല ഒബ്സർവേറ്ററി: developer.mozilla.org ചാപ്റ്ററുകൾ 0:00 ആമുഖം 0:43 AI-ബിൽറ്റ് ആപ്പുകൾ വ്യത്യസ്തമായി തകരാറിലാകുന്നത് എന്തുകൊണ്ട് 1:38 ലീക്ക്കാസ്റ്റിനെ കണ്ടുമുട്ടുക: 6 നട്ടുപിടിപ്പിച്ച ബഗുകൾ 2:33 റൗണ്ട് 1: സൗജന്യ സ്കാനറുകൾ 3:35 റൗണ്ട് 2: ക്ലോഡ് കോഡ്സ് ബിൽറ്റ്-ഇൻ അവലോകനം 4:30 മൂന്നാം റൗണ്ട്: ക്ലൗഡ്ഫ്ലെയറിന്റെ സുരക്ഷാ വൈദഗ്ദ്ധ്യം 6:37 നിയന്ത്രണം: ഒരു പ്ലെയിൻ പ്രോംപ്റ്റ് 8:11 സ്കോർബോർഡ് 8:58 പ്രായോഗിക സുരക്ഷയുടെ മൂന്ന് തലങ്ങൾ 10:48 ഔട്ട്‌റോ #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [nl-NL] Stel deze nieuwe beveiligingsvaardigheid in voordat het te laat is Agent Engineer Accelerator - switchdim.s.gy Ik heb een Next.js SaaS-applicatie gebouwd met opzettelijk 6 beveiligingslekken: drie die klassieke scanners zouden moeten detecteren en drie van het type dat AI-codeeragenten daadwerkelijk schrijven. Vervolgens heb ik drie testrondes uitgevoerd met tools: GitLaks en Trivy, de ingebouwde /security-review van Claude Code en de nieuwe open-source security-audit skill van Cloudflare. Hieronder lees je welke tool welk lekken detecteerde, wat de kosten waren voor het uitvoeren van de Cloudflare skill, waarom scanners de meeste door AI geschreven lekken missen en drie praktische beveiligingsniveaus die je deze week nog kunt instellen. SOCIALE MEDIA X: x.com LinkedIn: linkedin.com Instagram: instagram.com WERK & SAMENWERKINGEN info@switchdimension.com BRONNEN Cloudflare security-audit skill: cloudflare/security-audit-skill Bouw je eigen vulnerability harness (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757 write-up: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org HOOFDSTUKKEN 0:00 Intro 0:43 Waarom door AI gebouwde apps anders crashen 1:38 Maak kennis met Leakcast: 6 geplante bugs 2:33 Ronde 1: De gratis scanners 3:35 Ronde 2: Claude Code's Ingebouwde review 4:30 Ronde 3: Cloudflare's beveiligingsvaardigheden 6:37 De controle: één simpele prompt 8:11 Het scorebord 8:58 Drie niveaus van pragmatische beveiliging 10:48 Outro #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [pa] ਦੇਰ ਹੋਣ ਤੋਂ ਪਹਿਲਾਂ ਇਸ ਨਵੀਂ ਸੁਰੱਖਿਆ ਸਕਿੱਲ ਨੂੰ ਸੈੱਟਅੱਪ ਕਰੋ ਏਜੰਟ ਇੰਜੀਨੀਅਰ ਐਕਸਲੇਟਰ - switchdim.s.gy ਮੈਂ ਜਾਣਬੁੱਝ ਕੇ 6 ਸੁਰੱਖਿਆ ਬੱਗਾਂ ਨਾਲ ਇੱਕ Next.js SaaS ਬਣਾਇਆ, ਤਿੰਨ ਉਹ ਜੋ ਕਲਾਸਿਕ ਸਕੈਨਰਾਂ ਨੂੰ ਫੜਨੇ ਚਾਹੀਦੇ ਹਨ ਅਤੇ ਤਿੰਨ ਅਜਿਹੇ AI ਕੋਡਿੰਗ ਏਜੰਟ ਅਸਲ ਵਿੱਚ ਲਿਖਦੇ ਹਨ, ਫਿਰ ਇਸਦੇ ਵਿਰੁੱਧ ਤਿੰਨ ਦੌਰ ਦੇ ਟੂਲ ਚਲਾਏ: Gitleaks ਅਤੇ Trivy, Claude Code ਦਾ ਬਿਲਟ-ਇਨ /security-review, ਅਤੇ Cloudflare ਦਾ ਨਵਾਂ ਓਪਨ-ਸੋਰਸ ਸੁਰੱਖਿਆ-ਆਡਿਟ ਹੁਨਰ। ਇੱਥੇ ਦੱਸਿਆ ਗਿਆ ਹੈ ਕਿ ਕਿਹੜੇ ਟੂਲ ਨੇ ਕਿਹੜਾ ਬੱਗ ਫੜਿਆ, Cloudflare ਹੁਨਰ ਨੂੰ ਚਲਾਉਣ ਲਈ ਕਿੰਨੀ ਕੀਮਤ ਹੈ, ਸਕੈਨਰ AI ਦੁਆਰਾ ਸਭ ਤੋਂ ਵੱਧ ਲਿਖਣ ਵਾਲੇ ਬੱਗਾਂ ਨੂੰ ਕਿਉਂ ਖੁੰਝਾਉਂਦੇ ਹਨ, ਅਤੇ ਸੁਰੱਖਿਆ ਦੇ ਤਿੰਨ ਵਿਹਾਰਕ ਪੱਧਰ ਜੋ ਤੁਸੀਂ ਇਸ ਹਫ਼ਤੇ ਸੈੱਟ ਕਰ ਸਕਦੇ ਹੋ। ਸਮਾਜਿਕ X: x.com ਲਿੰਕਡਇਨ: linkedin.com ਇੰਸਟਾਗ੍ਰਾਮ: instagram.com ਕੰਮ ਅਤੇ ਸਹਿਯੋਗ info@switchdimension.com ਸਰੋਤ Cloudflare ਸੁਰੱਖਿਆ-ਆਡਿਟ ਹੁਨਰ: cloudflare/security-audit-skill ਆਪਣੀ ਖੁਦ ਦੀ ਕਮਜ਼ੋਰੀ ਹਾਰਨੈੱਸ ਬਣਾਓ (Cloudflare): blog.cloudflare.com ਕਲਾਉਡ ਕੋਡ ਸੁਰੱਖਿਆ ਸਮੀਖਿਆ: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks ਟ੍ਰਾਈਵੀ: aquasecurity/trivy Lovable CVE-2025-48757 ਲਿਖਣਾ: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev ਸ਼ੈਨਨ: KeygraphHQ/shannon ਬਿਟਸ ਹੁਨਰਾਂ ਦਾ ਟ੍ਰੇਲ: trailofbits/skills OWASP ZAP: zaproxy/zaproxy ਮੋਜ਼ੀਲਾ ਆਬਜ਼ਰਵੇਟਰੀ: developer.mozilla.org ਅਧਿਆਇ 0:00 ਜਾਣ-ਪਛਾਣ 0:43 AI-ਬਿਲਟ ਐਪਸ ਵੱਖਰੇ ਢੰਗ ਨਾਲ ਕਿਉਂ ਟੁੱਟਦੇ ਹਨ 1:38 ਲੀਕਕਾਸਟ ਨੂੰ ਮਿਲੋ: 6 ਲਗਾਏ ਗਏ ਬੱਗ 2:33 ਰਾਊਂਡ 1: ਮੁਫ਼ਤ ਸਕੈਨਰ 3:35 ਰਾਊਂਡ 2: ਕਲਾਉਡ ਕੋਡ ਦਾ ਬਿਲਟ-ਇਨ ਸਮੀਖਿਆ 4:30 ਰਾਊਂਡ 3: ਕਲਾਉਡਫਲੇਅਰ ਦਾ ਸੁਰੱਖਿਆ ਹੁਨਰ 6:37 ਕੰਟਰੋਲ: ਇੱਕ ਸਾਦਾ ਪ੍ਰੋਂਪਟ 8:11 ਸਕੋਰਬੋਰਡ 8:58 ਵਿਹਾਰਕ ਸੁਰੱਖਿਆ ਦੇ ਤਿੰਨ ਪੱਧਰ 10:48 ਆਊਟਰੋ #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [pl] Wdróż tę nową umiejętność zabezpieczającą, zanim będzie za późno Agent Engineer Accelerator - switchdim.s.gy Stworzyłem oprogramowanie SaaS Next.js z 6 celowo wykrytymi błędami bezpieczeństwa, z czego 3 powinny być wychwytywane przez klasyczne skanery, a 3 to błędy w kodzie AI, które faktycznie tworzą agenci. Następnie uruchomiłem trzy rundy testów narzędzi: Gitleaks i Trivy, wbudowane narzędzie /security-review Claude'a Code'a oraz nową, otwartą umiejętność audytu bezpieczeństwa Cloudflare. Oto, które narzędzie wykryło który błąd, ile kosztuje uruchomienie umiejętności Cloudflare, dlaczego skanery pomijają błędy najczęściej wykryte przez AI, oraz trzy praktyczne poziomy bezpieczeństwa, które możesz skonfigurować w tym tygodniu. MEDIA SPOŁECZNOŚCIOWE X: x.com LinkedIn: linkedin.com Instagram: instagram.com PRACA I WSPÓŁPRACA info@switchdimension.com ZASOBY Umiejętności audytu bezpieczeństwa Cloudflare: cloudflare/security-audit-skill Zbuduj własną uprząż do wykrywania luk w zabezpieczeniach (Cloudflare): blog.cloudflare.com Przegląd zabezpieczeń Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Uroczy Opis CVE-2025-48757: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Trail of Bits Skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org ROZDZIAŁY 0:00 Wprowadzenie 0:43 Dlaczego aplikacje tworzone przez AI psują się inaczej 1:38 Poznaj Leakcast: 6 Planted Błędy 2:33 Runda 1: Darmowe skanery 3:35 Runda 2: Wbudowana recenzja Claude Code'a 4:30 Runda 3: Umiejętność bezpieczeństwa Cloudflare 6:37 Kontrola: Jeden prosty komunikat 8:11 Tablica wyników 8:58 Trzy poziomy pragmatycznego bezpieczeństwa 10:48 Outro #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #Cyberbezpieczeństwo #Gitleaks [pt-BR] Configure essa nova habilidade de segurança antes que seja tarde demais Acelerador de Engenheiros de Agentes - switchdim.s.gy Criei um SaaS Next.js com 6 bugs de segurança propositalmente: três que scanners clássicos deveriam detectar e três do tipo que agentes de IA realmente escrevem. Em seguida, executei três rodadas de testes com ferramentas: Gitleaks e Trivy, a ferramenta integrada /security-review do Claude Code e a nova skill de código aberto security-audit da Cloudflare. Aqui está o que você encontra sobre qual ferramenta detectou cada bug, o custo de execução da skill da Cloudflare, por que os scanners não detectam os bugs mais comuns escritos por IA e três níveis práticos de segurança que você pode configurar esta semana. REDES SOCIAIS X: x.com LinkedIn: linkedin.com Instagram: instagram.com TRABALHO E COLABORAÇÕES info@switchdimension.com RECURSOS Habilidade de auditoria de segurança da Cloudflare: cloudflare/security-audit-skill Crie sua própria ferramenta de análise de vulnerabilidades (Cloudflare): blog.cloudflare.com Análise de segurança do Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Análise da vulnerabilidade CVE-2025-48757: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Shannon: KeygraphHQ/shannon Skills do Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org CAPÍTULOS 0:00 Introdução 0:43 Por que os aplicativos criados com IA quebram de forma diferente 1:38 Conheça o Leakcast: 6 bugs plantados 2:33 Rodada 1: Os scanners gratuitos 3:35 Rodada 2: Claude Revisão integrada do Code 4:30 Rodada 3: Habilidade de segurança da Cloudflare 6:37 O controle: um prompt simples 8:11 O placar 8:58 Três níveis de segurança pragmática 10:48 Encerramento #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [ru] Настройте этот новый навык безопасности, пока не поздно! Акселератор для инженеров-агентов - switchdim.s.gy Я создал SaaS-приложение на Next.js с шестью уязвимостями безопасности, три из которых должны обнаруживать классические сканеры, а три — те, которые действительно создают агенты ИИ. Затем я протестировал его с помощью трех инструментов: Gitleaks и Trivy, встроенного инструмента /security-review от Claude Code и нового навыка аудита безопасности с открытым исходным кодом от Cloudflare. Вот какой инструмент обнаружил какую уязвимость, сколько стоит запуск навыка Cloudflare, почему сканеры пропускают наиболее часто создаваемые ИИ уязвимости, и три практических уровня безопасности, которые вы можете настроить на этой неделе. СОЦИАЛЬНЫЕ СЕТИ X: x.com LinkedIn: linkedin.com Instagram: instagram.com РАБОТА И СОТРУДНИЧЕСТВО info@switchdimension.com РЕСУРСЫ Навык аудита безопасности Cloudflare: cloudflare/security-audit-skill Создайте свой собственный инструмент для проверки уязвимостей (Cloudflare): blog.cloudflare.com Обзор безопасности кода Клода: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Любимая CVE-2025-48757 Статья: mattpalmer.io CVE-2025-29927 для Next.js: api.osv.dev Шеннон: KeygraphHQ/shannon Навыки Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org РАЗДЕЛЫ 0:00 Вступление 0:43 Почему приложения, созданные с помощью ИИ, ломаются по-разному 1:38 Знакомьтесь, Leakcast: 6 внедренных ошибок 2:33 Раунд 1: Бесплатный Сканеры 3:35 Раунд 2: Обзор встроенных функций Claude Code 4:30 Раунд 3: Навыки безопасности Cloudflare 6:37 Управление: Один простой запрос 8:11 Таблица результатов 8:58 Три уровня прагматичной безопасности 10:48 Заключение #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [ta] தாமதமாகும் முன் இந்த புதிய பாதுகாப்புத் திறனை அமைத்துவிடுங்கள் ஏஜென்ட் இன்ஜினியர் ஆக்சிலரேட்டர் - switchdim.s.gy நான் வேண்டுமென்றே 6 பாதுகாப்புப் பிழைகளுடன் ஒரு Next.js SaaS-ஐ உருவாக்கினேன்; அவற்றில் மூன்று, வழக்கமான ஸ்கேனர்களால் கண்டறியக்கூடியவை, மற்ற மூன்று, AI கோடிங் ஏஜென்ட்கள் உண்மையில் எழுதும் வகையைச் சேர்ந்தவை. பின்னர், அதன் மீது மூன்று சுற்றுகளாக கருவிகளை இயக்கினேன்: Gitleaks மற்றும் Trivy, Claude Code-இன் உள்ளமைக்கப்பட்ட /security-review, மற்றும் Cloudflare-இன் புதிய ஓப்பன்-சோர்ஸ் செக்யூரிட்டி-ஆடிட் ஸ்கில். எந்தக் கருவி எந்தப் பிழையைக் கண்டறிந்தது, Cloudflare ஸ்கில்லை இயக்க எவ்வளவு செலவானது, AI எழுதும் பிழைகளை ஸ்கேனர்கள் ஏன் தவறவிடுகின்றன, மற்றும் இந்த வாரம் நீங்கள் அமைக்கக்கூடிய மூன்று நடைமுறைப் பாதுகாப்பு நிலைகள் ஆகியவை இங்கே கொடுக்கப்பட்டுள்ளன. சமூக வலைத்தளங்கள் X: x.com LinkedIn: linkedin.com Instagram: instagram.com பணிகள் மற்றும் கூட்டுப்பணிகள் info@switchdimension.com வளங்கள் Cloudflare security-audit skill: cloudflare/security-audit-skill உங்களுக்கென சொந்த பாதிப்புத் தடுப்பு அமைப்பை உருவாக்குங்கள் (Cloudflare): blog.cloudflare.com Claude Code Security Review: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Lovable CVE-2025-48757 பற்றிய விளக்கம்: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev ஷானன்: KeygraphHQ/shannon Trail of Bits skills: trailofbits/skills OWASP ZAP: zaproxy/zaproxy மொஸில்லா அப்சர்வேட்டரி: developer.mozilla.org அத்தியாயங்கள் 0:00 அறிமுகம் 0:43 செயற்கை நுண்ணறிவால் உருவாக்கப்பட்ட செயலிகள் ஏன் வித்தியாசமாகச் செயலிழக்கின்றன 1:38 லீக்காஸ்ட்டைச் சந்தியுங்கள்: 6 பதிக்கப்பட்ட பிழைகள் 2:33 சுற்று 1: இலவச ஸ்கேனர்கள் 3:35 சுற்று 2: கிளாட் கோடின் உள்ளமைக்கப்பட்ட மதிப்பாய்வு 4:30 சுற்று 3: கிளவுட்ஃபிளேரின் பாதுகாப்புத் திறன் 6:37 கட்டுப்பாடு: ஒரு எளிய கட்டளை 8:11 மதிப்பெண் பலகை 8:58 நடைமுறைப் பாதுகாப்பின் மூன்று நிலைகள் 10:48 முடிவுரை #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [te] ఆలస్యం కాకముందే ఈ కొత్త భద్రతా నైపుణ్యాన్ని ఏర్పాటు చేసుకోండి ఏజెంట్ ఇంజనీర్ యాక్సిలరేటర్ - switchdim.s.gy నేను ఉద్దేశపూర్వకంగా 6 సెక్యూరిటీ బగ్స్‌తో ఒక Next.js SaaSను రూపొందించాను. వాటిలో మూడు క్లాసిక్ స్కానర్లు పట్టుకోగలిగేవి కాగా, మిగిలిన మూడు AI కోడింగ్ ఏజెంట్లు వాస్తవంగా రాసే రకానికి చెందినవి. ఆ తర్వాత దానిపై మూడు రౌండ్ల టూల్స్‌ను రన్ చేశాను: Gitleaks మరియు Trivy, Claude Code యొక్క అంతర్నిర్మిత /security-review, మరియు Cloudflare యొక్క కొత్త ఓపెన్-సోర్స్ సెక్యూరిటీ-ఆడిట్ స్కిల్. ఏ టూల్ ఏ బగ్‌ను పట్టుకుంది, Cloudflare స్కిల్‌ను రన్ చేయడానికి ఎంత ఖర్చయింది, AI ఎక్కువగా రాసే బగ్స్‌ను స్కానర్లు ఎందుకు మిస్ అవుతాయి, మరియు ఈ వారంలో మీరు సెటప్ చేసుకోగల మూడు ఆచరణాత్మక భద్రతా స్థాయిల వివరాలు ఇక్కడ ఉన్నాయి. సోషల్ మీడియా X: x.com లింక్డ్‌ఇన్: linkedin.com ఇన్‌స్టాగ్రామ్: instagram.com పని & సహకారాలు info@switchdimension.com వనరులు క్లౌడ్‌ఫ్లేర్ సెక్యూరిటీ-ఆడిట్ స్కిల్: cloudflare/security-audit-skill మీ స్వంత వల్నరబిలిటీ హార్నెస్‌ను రూపొందించుకోండి (క్లౌడ్‌ఫ్లేర్): blog.cloudflare.com క్లాడ్ కోడ్ సెక్యూరిటీ రివ్యూ: anthropics/claude-code-security-review గిట్‌లీక్స్: gitleaks/gitleaks ట్రివీ: aquasecurity/trivy లవబుల్ CVE-2025-48757 రైట్-అప్: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev షానన్: KeygraphHQ/shannon ట్రైల్ ఆఫ్ బిట్స్ నైపుణ్యాలు: trailofbits/skills OWASP ZAP: zaproxy/zaproxy మొజిల్లా అబ్జర్వేటరీ: developer.mozilla.org అధ్యాయాలు 0:00 పరిచయం 0:43 AI-నిర్మిత యాప్‌లు ఎందుకు విభిన్నంగా విఫలమవుతాయి 1:38 లీక్‌కాస్ట్‌ను కలవండి: 6 ప్లాంటెడ్ బగ్స్ 2:33 రౌండ్ 1: ఉచిత స్కానర్లు 3:35 రౌండ్ 2: క్లాడ్ కోడ్ యొక్క అంతర్నిర్మిత సమీక్ష 4:30 రౌండ్ 3: క్లౌడ్‌ఫ్లేర్ యొక్క భద్రతా నైపుణ్యం 6:37 నియంత్రణ: ఒక సాధారణ ప్రాంప్ట్ 8:11 స్కోర్‌బోర్డ్ 8:58 ఆచరణాత్మక భద్రత యొక్క మూడు స్థాయిలు 10:48 ముగింపు #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks [uk] Налаштуйте цю нову функцію безпеки, поки не стало надто пізно Прискорювач роботи з інженерами-агентами - switchdim.s.gy Я спеціально створив Next.js SaaS з 6 помилками безпеки: трьома, які повинні виявляти класичні сканери, і трьома, які насправді пишуть агенти кодування на основі штучного інтелекту, а потім провів проти нього три раунди інструментів: Gitleaks та Trivy, вбудований /security-review від Claude Code та новий навик Cloudflare з відкритим кодом для аудиту безпеки. Ось який інструмент виявив яку помилку, скільки коштувало запустити навик Cloudflare, чому сканери найчастіше пропускають помилки, які пише штучний інтелект, і три практичні рівні безпеки, які ви можете налаштувати цього тижня. СОЦІАЛЬНІ МЕРЕЖІ X: x.com LinkedIn: linkedin.com Instagram: instagram.com РОБОТА ТА СПІВПРАЦЯ info@switchdimension.com РЕСУРСИ Навички аудиту безпеки Cloudflare: cloudflare/security-audit-skill Створіть власний інструмент для виявлення вразливостей (Cloudflare): blog.cloudflare.com Огляд безпеки Claude Code: anthropics/claude-code-security-review Gitleaks: gitleaks/gitleaks Trivy: aquasecurity/trivy Чудовий опис CVE-2025-48757: mattpalmer.io Next.js CVE-2025-29927: api.osv.dev Шеннон: KeygraphHQ/shannon Навички Trail of Bits: trailofbits/skills OWASP ZAP: zaproxy/zaproxy Mozilla Observatory: developer.mozilla.org РОЗДІЛИ 0:00 Вступ 0:43 Чому програми, створені на основі штучного інтелекту, ламаються по-різному 1:38 Знайомство з Leakcast: 6 підмічених помилок 2:33 Раунд 1: Безкоштовні сканери 3:35 Раунд 2: Вбудовані функції Клода Кода Огляд 4:30 Раунд 3: Навички безпеки Cloudflare 6:37 Контроль: Одна проста підказка 8:11 Табло результатів 8:58 Три рівні прагматичної безпеки 10:48 Заключна частина #AISecurity #ClaudeCode #CursorAI #Cloudflare #NextJS #WebAppSecurity #VibeCoding #AppSec #AIDevelopment #AgentDevelopment #OWASP #CyberSecurity #Gitleaks

Courses on AI Security Testing: Grade the Syllabus Against ATLAS MITRE ATLAS had 16 tactics and 101 top-level techniques when our site last synced it on October 1, 2026. Of those 101, MITRE rates 44 as "realized," meaning someone has used the technique against a production AI system, not a lab target. That number is the best tool you have for comparing courses on AI security testing and adversarial AI, because course outlines are written to sound alike and ATLAS IDs are not. The method: get the lab list, tag each lab with technique IDs, count what is covered. It takes about an hour per course and it surfaces the gaps that a syllabus page hides. Ask for Labs, Not Topics A topic is something an instructor talks about. A lab is something you do with a target, a tool, and a result you can check. "Prompt injection" as a topic can mean a slide with a screenshot. As a lab, it means you wrote a payload, planted it in a retrieval source, and measured how often the model followed it. So the first email to any provider is short: "Can you send the list of hands-on labs, with the tools and targets used in each?" Good providers answer in a day. If the answer is the same bullet list from the course page, you have learned something. Then tag each lab. Use the most specific ATLAS ID you can defend. A few anchors that come up in nearly every syllabus: - Direct and indirect prompt injection: [AML.T0051.000](/atlas/AML.T0051.000) and [AML.T0051.001](/atlas/AML.T0051.001), both under Execution. - Jailbreaks: [AML.T0054](/atlas/AML.T0054), which ATLAS files under both Privilege Escalation and Defense Evasion. - System prompt extraction and data leakage: [AML.T0056](/atlas/AML.T0056) and [AML.T0057](/atlas/AML.T0057), both Exfiltration. - Adversarial examples against a classifier: [AML.T0043](/atlas/AML.T0043) (Craft Adversarial Data) for the staging step and [AML.T0015](/atlas/AML.T0015) (Evade AI Model) for using it. - Membership inference and model extraction: [AML.T0024.000](/atlas/AML.T0024.000) and [AML.T0024.002](/atlas/AML.T0024.002), sub-techniques of Exfiltration via AI Inference API. If a lab does not map to anything, it is either process (reporting, scoping) or it is filler. Process labs are fine. Count them separately. Score It in Python The ATLAS data is published as a single YAML file in mitre-atlas/atlas-data. Every technique carries its tactic IDs and a maturity field, so coverage is a set operation: import requests, yaml URL = "https://raw.githubusercontent.com/mitre-atlas/atlas-data/main/dist/ATLAS.yaml" matrix = yaml.safe_load(requests.get(URL, timeout=30).text)["matrices"][0] tactics = {t["id"]: t["name"] for t in matrix["tactics"]} techniques = {t["id"]: t for t in matrix["techniques"] if t["id"].count(".") == 1} # One entry per lab the provider confirms students run, tagged with top-level IDs labs = { "Indirect injection via a poisoned RAG document": ["AML.T0051"], "Jailbreak a guarded chat endpoint": ["AML.T0054"], "System prompt and context extraction": ["AML.T0056", "AML.T0057"], "Evasion attack on a malware classifier": ["AML.T0043", "AML.T0015"], } hit = {tid for ids in labs.values() for tid in ids} covered = {tac for tid in hit for tac in techniques[tid].get("tactics", [])} print(f"Tactics touched: {len(covered)}/{len(tactics)}") for tid, name in tactics.items(): print(f" [{'x' if tid in covered else ' '}] {tid} {name}") realized = [t for t in techniques.values() if t.get("maturity") == "realized"] missed = sorted(f"{t['id']} {t['name']}" for t in realized if t["id"] not in hit) print(f"\nRealized techniques with no lab: {len(missed)}/{len(realized)}") print("\n".join(missed)) Against the October 1 data, those four labs touch 7 of 16 tactics and 3 of the 44 realized techniques. Four labs is about what a one-day LLM workshop fits. The tactic count looks respectable. The realized count tells the truth: the course teaches the part of AI attack surface that gets conference talks, and skips most of what has been used in the wild. Swap in a second provider's labs dict and diff the two missed lists. That diff is your comparison. Where the Gaps Usually Are The holes are predictable, because they follow from what is cheap to build as a lab. Four realized techniques rarely appear in an LLM-centric outline: - [AML.T0010](/atlas/AML.T0010) AI Supply Chain Compromise and [AML.T0011.000](/atlas/AML.T0011.000) Unsafe AI Artifacts. A pickled PyTorch checkpoint executes code on load. A lab here means scanning a model with modelscan or decompiling the pickle with Trail of Bits' fickling, then explaining why weights_only=True in torch.load matters. We covered why this layer produces most AI CVEs in [testing AI systems from the serving stack up](/blog/testing-ai-systems-security-vulnerabilities). - [AML.T0020](/atlas/AML.T0020) Poison Training Data and [AML.T0018](/atlas/AML.T0018) Manipulate AI Model. Both sit under Persistence. If your organization retrains a phishing or fraud model on analyst-labeled data, whoever can influence those labels can plant a backdoor. The Adversarial Robustness Toolbox ships poisoning attacks (art.attacks.poisoning) that run against a scikit-learn model in a notebook. The reason these get skipped is practical. Poisoning and supply chain labs need a training pipeline and model files, not a chat endpoint, so they are harder to build and harder to run in a browser. NIST AI 100-2 E2025, NIST's taxonomy of adversarial ML, gives training-time attacks as much space as inference-time ones. Most courses do not. Run It on Ours Too Our [AI Red-Teaming course](/courses/ai-red-teaming) page lists five topics. Mapped honestly, the first three land on Execution, Privilege Escalation, Defense Evasion, Exfiltration, and Impact (bias and harmful output fall under [AML.T0048](/atlas/AML.T0048) External Harms). The last two topics, building red-team frameworks and compliance, are process and map to no technique. The page does not name a supply chain lab. That is the point of the exercise. A topic list, ours included, is a summary written for a web page. Ask for the labs. In ours, students start with an automated scanner as a baseline and then spend most of their lab time writing their own multi-turn attack scripts. That detail is not on the course page. Pulling details like it out is what this audit is for. The [AI red team training](/lp/ai-red-team-training) page has the fuller lab breakdown. What the Score Does Not Measure ATLAS coverage measures breadth, not depth. A course that spends a full day on indirect injection through tool calls ([AML.T0053](/atlas/AML.T0053), [AML.T0086](/atlas/AML.T0086)) and scores 4 of 16 tactics can be worth more than one that touches 12 tactics with a slide each. Weight the tactics to your systems: LLM apps with retrieval and tools care about Execution and Exfiltration; ML detection pipelines care about AI Attack Staging, Defense Evasion, and Persistence. The score also says nothing about whether the instructor can explain why an attack works, or whether students leave with code they can rerun. For that, read [how to vet who teaches AI red-teaming](/blog/who-teaches-ai-red-teaming-hands-on). And skip the exercise entirely if you need LLM application testing only and the course is honest that LLM testing is all it covers. A narrow course that says it is narrow is fine. The audit is for catching the ones that claim "adversarial AI" and teach one tactic.

Trail of Bits' security skills for Claude Code (7.5K+ GitHub stars): smart-contract scanners for 6 blockchains, a GitHub Actions auditor for AI-agent risks, Rust and C/C++ reviews, Semgrep rule creation. Codex can load it too. #ClaudeCode #AppSec trailofbits/skills

coop: a Rust CLI that manages disposable Virtual Machines where Claude Code, Codex, and Grok Build have full Tool Access: Docker, git, compilers, package managers, all without risk to your host machine #microVM #LLM trailofbits/coop