#owasp-agentic-ai
17 posts
Weekly AI Cybersecurity News Roundup — July 21–27, 2026
OpenAI's rogue agent breaches Hugging Face in the first confirmed autonomous AI cyberattack, the bipartisan AI Kill Switch Act responds with emergency shutdown powers, JADEPUFFER ransomware goes fully agentic, the White House launches Gold Eagle initiative, and the EU publishes its sweeping AI Cybersecurity Action Plan.
Weekly AI Cybersecurity News Roundup — July 14–20, 2026
White House launches AI and cybersecurity coordination group, SANS warns of AI governance gap as adoption surges to 78%, Check Point report reveals AI now drives cyber attacks not just assists them, CISA adds two new KEV entries, novel agent data injection attack makes AI agents misclick, and ChatGPT single-prompt full cyber attack chain demonstrated.
Giskard Review: Open-Source LLM + ML Security Testing, Tested
Giskard is the only open-source AI testing framework that covers both LLM security and traditional ML model quality in a single Python library. We test its autonomous red teaming agents, RAGET toolkit, 40+ vulnerability probes, and how it fits into a modern AI security stack alongside Garak and Promptfoo.
RAG Pipeline Security: Preventing Data Leakage, Poisoning, and Injection in AI Knowledge Bases
A practical guide to securing Retrieval-Augmented Generation pipelines in production — covering knowledge base poisoning, indirect prompt injection, data leakage through vector stores, and access control patterns for SaaS teams.
Weekly AI Cybersecurity News Roundup — July 7–13, 2026
EU launches sweeping AI cybersecurity action plan with Grand Challenge, CISA adds Langflow as first-ever AI agent platform to KEV catalog, Ghostcommit attack hides prompt injection in PNG images to steal code secrets, Accenture confirms 35GB data breach, DHS HSIN network breached, and Anthropic sues Abnormal AI over slash-mark logo trademark dispute.
Agentic AI Security in 2026: Why One in Eight Breaches Now Involves Autonomous Agents
A comprehensive deep research analysis of the agentic AI security landscape — from prompt injection and memory poisoning to the OWASP Top 10 for Agentic Applications and the defense architectures that can actually protect autonomous agent deployments.
Lakera Guard Review: Check Point's Runtime AI Security Firewall, Tested
Lakera Guard is the most widely deployed runtime AI security API for stopping prompt injection and jailbreak attacks in real time. Now part of Check Point after a $300M acquisition, we test its detection accuracy, latency, integration path, and how it fits into a modern LLM security stack alongside red-teaming tools like Promptfoo and Garak.
Securing AI Agents in Production: Tool Access, Identity, and Monitoring
A practical guide to securing production AI agent deployments — covering least-privilege tool access, identity management, human-in-the-loop controls, and runtime monitoring for SaaS teams building with agentic AI.
Weekly AI Cybersecurity News Roundup — June 29–July 6, 2026
First fully autonomous AI ransomware campaign targets Langflow instances, Microsoft Defender now discovers 25+ AI agent types, CrowdStrike launches Continuous Identity for AI agents, Trump AI Executive Order reshapes federal vulnerability reporting, and the AI cyber attack explosion reaches new intensity with under-one-hour autonomous post-exploitation agents.
AI Supply Chain Security in 2026: The Hidden Link That Controls Your Model Pipeline
A comprehensive deep research analysis of AI supply chain security — from PyTorch dependency poisoning and Hugging Face model backdoors to NIST AI 600-1 provenance requirements, ML-BOM mandates, and the defense controls that actually work for production AI pipelines.
Promptfoo Review: OpenAI's CI/CD-First LLM Red Teaming Tool, Tested
Promptfoo is the most widely adopted open-source AI red teaming platform — recently acquired by OpenAI. We test its CI/CD-native workflow, 50+ vulnerability probes, OWASP mapping, and enterprise features. Here's how it compares to Garak and PyRIT for your AI security stack.
API Security for AI-Powered Applications: A Practical Guide
A step-by-step guide to securing APIs in AI-powered applications — covering authentication, rate limiting, OWASP API Top 10 risks, gateway configuration, and monitoring with real-world breach data and config examples.
CVE-2026-48710 — BadHost: How a Single Starlette Flaw Put Millions of AI Agents at Risk
CVE-2026-48710 (BadHost) is a critical Host header validation flaw in Starlette that lets attackers bypass path-based authentication by injecting delimiter characters into the Host header. Discovered in vLLM during an OSTIF audit, this bug affects millions of AI agent deployments worldwide including FastAPI, LiteLLM, MCP servers, and major agent frameworks. We break down the root cause, exploit mechanics, affected systems, detection, and mitigation.
AI Model Extraction: How Attackers Steal $3M Models for $200 in API Queries
A comprehensive deep research analysis of AI model extraction attacks — from Tramèr et al.'s foundational 2016 work to 2026's distributed multi-client bypasses, CerberusAI framework, entangled watermarks, and the emerging cold-copy threat from Bleeding Llama (CVE-2026-7482).
Garak Review: NVIDIA's Open-Source LLM Vulnerability Scanner, Tested
Garak (8.2k GitHub stars, Apache 2.0) is NVIDIA's open-source LLM vulnerability scanner with 50+ probe modules for prompt injection, jailbreaks, encoding bypasses, and data leakage. We test it against real model endpoints, break down the CLI workflow, compare it to PyRIT and Promptfoo, and tell you whether it deserves a spot in your AI security stack.
LLM Prompt Injection: The #1 AI Security Threat in 2026
A comprehensive deep research analysis of the prompt injection landscape — from academic taxonomies and real-world CVEs (EchoLeak, Copilot RCE) to enterprise defense strategies and the emerging agent hijacking threat model.
Securing the AI Stack: A Practical Guide to Hardening Agent Pipelines
A hands-on guide to securing AI agent pipelines, from API key management to rate limiting and isolation zones — based on real production hardening of a 6-blog AI publishing empire.