A syllabus for the adversarial life of large language models

What it does

This is a curated awesome-list that catalogs papers, tools, benchmarks, and articles about LLM security. It sorts research into tidy taxonomies: white-box attacks, black-box attacks, backdoors, fingerprinting, defenses, and platform security. Think of it as a living literature review maintained by pull request.

The interesting bit

The list gives equal shelf space to offense and defense, cataloguing everything from visual adversarial jailbreaks to multi-agent defense systems. The maintainers also share PDFs via Moonlight, pairing each entry with a summary so you can decide whether the full academic slog is worth your time.

Key highlights

• Covers the full adversarial lifecycle, from visual adversarial examples and audio injection to backdoor triggers and model fingerprinting.
• Heavy emphasis on jailbreak research, including cipher-based stealth attacks, persona modulation, and multi-modal prompt injection.
• Surfaces defense papers proposing self-examination, random-mask filtering, and multi-agent LLM defenses.
• Links directly to repositories, datasets, and Hugging Face assets where available.
• Includes a dedicated section for benchmarks and practical tools, not just theoretical papers.

Verdict

Worth bookmarking if you ship LLM features and need a crash course on how they break. Skip it if you want a hands-on framework; this is a reading list, not a toolkit.