Secure By Dezign

Episode 41: Federated Learning Poisoning Weaponizing Collaborative Ai

Episode 42: Attacking Ml Api Gateways Behavioral Drift Model Poisoning

Episode 39: Clean Label Poisoning Invisible Training Data Attack

Episode 40: Backdoor Attacks Trojaned Neural Networks

Episode 38: Payload Splitting Bypassing Ai Filters

How attackers weaponize conversational context to make LLMs forget their guardrails across multi-turn interactions — and how to build defenses that persist across the full conversation window.

Exploiting the gap between human-readable text and machine tokenization to bypass every filter you've built. Covers homoglyph attacks, whitespace injection, and tokenizer-aware defenses.

Your airgapped playground for prompt injection, jailbreaking, and system prompt extraction. No API keys, no rate limits, no excuses. Complete setup and attack walkthrough with Ollama.

Your trusted knowledge base is an injection surface — here's how attackers exploit it. Hands-on lab covering document poisoning, embedding manipulation, and retrieval hijacking.

Set up a local LLM agent with dangerous tools, then systematically exploit it with tool injection, privilege escalation, memory hijacking, and DoS. Full hands-on lab walkthrough.

When your image is the attack vector, every picture becomes a potential payload. Technical walkthrough of visual prompt injection against multimodal LLMs including GPT-4V and Gemini Vision.

Why your carefully aligned model is one clever prompt away from chaos. Covers DAN variants, many-shot jailbreaking, adversarial suffixes, and the cat-and-mouse dynamics of safety alignment.

A strategic allocation framework for securing AI systems while demonstrating ROI to the board — including tooling prioritization, build vs. buy decisions, and budget defense strategies.

When your LLM trusts external content, attackers don't need access to your users — they just need a webpage. Technical walkthrough of indirect prompt injection with real-world exploitation chains.

Your model is only as trustworthy as the data that built it. A practitioner's guide to defending every stage of the ML data pipeline — from ingestion to labeling to preprocessing.

The executive playbook for vetting AI suppliers in an era of opaque models and expanding attack surfaces. What contractual, technical, and audit controls CISOs must demand.

A practitioner's guide to implementing AI-SPM frameworks before regulatory mandates force your hand — covering emerging standards, tooling, and continuous posture assessment.

Technical strategies for embedding, detecting, and enforcing ownership claims in production ML systems — including robustness testing against removal attacks.

Building enforceable AI governance that survives first contact with production systems. Covers NIST AI RMF, EU AI Act, ISO 42001, and practical implementation strategies.

How adversaries weaponize natural language to compromise your most trusted AI systems. Comprehensive coverage of direct and indirect injection, RLHF evasion, and enterprise-grade defenses.

How adversaries weaponize the trust you place in pre-trained models, datasets, and ML dependencies. Covers backdoor injection, dataset poisoning, and dependency confusion in the ML pipeline.

When employees bypass security to use ChatGPT, Claude, and dozens of other AI tools, they're not just breaking policy — they're creating attack surfaces you can't see. Here's how to find and fix them.

Autonomous AI systems are rewriting the threat landscape. Deep-dive into agent hijacking, tool misuse, privilege escalation in multi-agent pipelines, and your survival guide for agentic deployments.

The distributed ML paradigm that promised privacy may be your biggest attack surface yet. Deep-dive into poisoning attacks against federated models, gradient inversion, and Byzantine fault tolerance.

Translating AI security threats into boardroom language that drives action and budget — with model answers to the toughest questions your board will ask in 2026.