OWASP GenAI Security Project ↗
Risk taxonomy and educational resources for LLM applications. DAI/ADAI address authority boundaries, effect scope and evidence; this does not assert certification or eliminate every LLM risk.
Related research
These links explain research relationships, not endorsements, partnerships, dependencies or benchmark results. Consult each project’s current documentation and license.
AgentDojo ↗
A dynamic environment for evaluating prompt-injection attacks and defenses in LLM agents. Relevant evaluation methodology; DAI is an LLM, and the manuscripts do not report AgentDojo evaluation or integration.
Prompt Injection Defenses ↗
A collection of defensive approaches, including least privilege and blast-radius reduction. Related to bounding the effect of untrusted model output; no universal prompt-injection immunity is claimed.
Awesome LLM Security ↗
Curated research on language-model security, including platform security. A discovery path for comparing research questions and evaluation assumptions.
Awesome AI Security ↗
Resources on AI security, governance and defensive systems. Useful context for the papers’ distinction between model output and execution authority.
Awesome LM Safety, Security and Privacy ↗
A research collection across safety, security and privacy. System-security research is adjacent to the envelope-and-evidence framework.