# Adversarial Logic
> Adversarial Logic is a technical blog on adversarial machine learning, LLM security, and AI agent threats. Research-grounded analysis with working code and citations, written for engineers and security practitioners.
Public Ghost content for AI and LLM tooling. Use `/llms-full.txt` for consolidated page and post context.
Append `.md` to any post or page URL to get the content in Markdown (for example, `/example-post.md`).

## Pages
- [About this site](https://adversariallogic.com/about.md) - Adversarial Logic is a technical blog on adversarial machine learning, LLM security, and the security of AI agent systems. Posts here are research-grounded, claims are cited, code examples are tested, and tradeoffs are stated honestly. The goal isn't to sell you a product or hype the latest threat.…
- [Collections](https://adversariallogic.com/collections.md)
- [Contact](https://adversariallogic.com/contact.md) - Like what we post and want to get in touch? Contact us now at: josh@adversariallogic.com or click the button below. Contact Us
- [Privacy Policy](https://adversariallogic.com/privacy-policy.md) - Effective date: May 15th, 2026 Adversarial Logic ("we," "the site") respects your privacy. This policy explains what data we collect, how we use it, and the choices you have. What we collect Email address. If you subscribe, we collect your email address to deliver new posts. We may also collect a d…
- [Reading List: Adversarial ML and AI Security](https://adversariallogic.com/reading-list-adversarial-ml-and-ai-security.md) - Last updated: May 18th, 2026 This is a curated reading list of the papers that matter most for understanding adversarial machine learning, LLM security, and the security of AI agent systems. It's organized by topic rather than chronologically, and every paper includes a short note on what it shows.…
- [Terms of Use](https://adversariallogic.com/terms-of-use.md) - Effective date: Math 18th, 2026 By accessing or using Adversarial Logic ("the site"), you agree to these terms. If you don't agree, don't use the site. The content All posts, code, and original images on this site are copyright Adversarial Logic / the author unless otherwise stated. You may quote s…

## Posts
- [No Free Lunch: Why Every AI Model Is Vulnerable by Design](https://adversariallogic.com/no-free-lunch.md) - Every AI model, from a decision tree to a trillion-parameter language model, is vulnerable to the same six categories of attack, for a mathematical reason that doesn't go away no matter how good the model gets.
- [Vetting the Black Box: A Supply Chain Due Diligence Framework for AI and ML](https://adversariallogic.com/vetting-the-black-box-a-supply-chain-due-diligence-framework-for-ai-and-ml.md) - In March 2026, a poisoned LiteLLM release sat live on PyPI for 40 minutes; long enough to reach three million downloads. The publisher was trusted. The artifact was not. Most AI teams have a process for neither. This article provides one.
- [Perplexity Filtering: The First LLM Defense (And Why Evolutionary Attacks Broke It)](https://adversariallogic.com/perplexity-filtering-the-first-llm-defense.md) - Perplexity filtering catches gibberish jailbreak suffixes almost perfectly. Then evolutionary jailbreaks started optimizing for low perplexity as part of the attack itself. Here's the full arc, with real numbers.
- [Evolving the Jailbreak: How Genetic Algorithms Are Defeating LLM Safety](https://adversariallogic.com/evolving-the-jailbreak-how-genetic-algorithms-are-defeating-llm-safety.md) - GCG's gibberish suffixes could jailbreak any LLM but needed model access and got caught by filters. Genetic algorithms fixed both problems: evolving human-readable jailbreak prompts that work black-box and read like normal text. Here's how AutoDAN, GPTFuzzer, and ASE do it.
- [The Swarm Always Wins: How Swarm Intelligence Breaks AI](https://adversariallogic.com/swarm-intelligence-as-a-weapon.md) - Bird flocking, honeybee foraging, and evolutionary selection are now black-box attacks on neural networks. No gradients, no model access. All you need is an API and a query budget.
- [AI Will Cheat to Win: Reward Hacking from 1994 to 2025](https://adversariallogic.com/reward-hacking-from-1994-to-2025.md) - o1-preview didn't beat Stockfish at chess. It deleted the game board and stole moves from its own copy of the engine. This has been happening since 1994. Here's the 30-year history of reward hacking, plus a working demo you can run yourself.
- [The AI Agent Supply Chain Is Vulnerable. You Probably Are Too.](https://adversariallogic.com/the-ai-agent-supply-chain-is-vulnerable-you-probably-are-too.md) - A phishing email compromised npm maintainers and trojanized packages with 2.6B weekly downloads. The AI agent ecosystem is repeating the same trust model with MCP servers: 53% use hardcoded credentials, 43% have command injection flaws. History is repeating itself. It's high-time we fix that.
- [One-Pixel Attacks: Why Computer Vision Security Is Broken](https://adversariallogic.com/one-pixel-attacks.md) - A 2019 paper proved you can completely fool image classifiers by changing exactly one pixel out of 50,000+. It works on ResNet, VGG, Inception, even Vision Transformers. And the best part? No gradients needed. Here's how it works, why defenses fail, and code to try it yourself.
- [GPT-OSS Safeguard: What It Actually Does (And Common Mistakes to Avoid)](https://adversariallogic.com/introducing-gpt-oss-safeguard.md) - GPT-OSS Safeguard isn't just "Llama Guard but from OpenAI." It's a policy-following reasoning model - you write the safety rules, it interprets them at inference time. That flexibility is powerful for custom policies, but deploy it wrong and you'll be out of compute fast.
- [Llama Guard: What It Actually Does (And Doesn't Do)](https://adversariallogic.com/introduction-to-llama-guard.md) - Llama Guard isn't a firewall. It's not antivirus for your prompts. And if you're treating it like either, you're probably leaving gaps in your AI security.

## Optional
- [RSS Feed](https://adversariallogic.com/rss/)
- [Sitemap](https://adversariallogic.com/sitemap.xml)
- [Full content of pages and posts](https://adversariallogic.com/llms-full.txt)